🔭 astrophysics

Fisher Forecasting for the DESC with $\texttt{Augur}$

This paper introduces $\texttt{Augur}$ , a computationally efficient Fisher forecasting tool developed by the Dark Energy Science Collaboration to evaluate cosmological modeling choices for the Vera C. Rubin Observatory's LSST survey, which has been validated against external codes and nested sampling methods.

Original authors: Paul Rogozenski, Sankarshana Srinivasan, Javier Sánchez, Nora Elisa Chisari, Arthur Loureiro, Marc Paterno, Rebekah Polen, Heather Prince, Biancamaria Sersante, Anže Slosar, Sandro Vitenti, Carlos Gar

Published 2026-08-05

📖 4 min read☕ Coffee break read

CC BY 4.0

Original authors: Paul Rogozenski, Sankarshana Srinivasan, Javier Sánchez, Nora Elisa Chisari, Arthur Loureiro, Marc Paterno, Rebekah Polen, Heather Prince, Biancamaria Sersante, Anže Slosar, Sandro Vitenti, Carlos García-García, Eric Gawiser, Christos Georgiou, C. Danielle Leonard, Ayan Mitra, Jeremy Neveu, 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. Most of what's in it isn't the water we can see (stars and planets), but two mysterious, invisible ingredients: "dark matter," which acts like a cosmic glue holding galaxies together, and "dark energy," a strange force pushing the universe apart faster and faster. For decades, scientists have been trying to measure exactly how much of each exists and how they behave. To do this, they build massive "telescopes" that aren't just one big mirror, but entire surveys that scan the sky for years, looking at how light bends around invisible mass (a trick called "gravitational lensing") and how galaxies cluster together. The goal is to solve the mystery of why the universe is accelerating and to see if our current rules of physics are actually broken.

But before these giant telescopes even start collecting real data, scientists need to know: "Will our plan work?" They need to run a simulation to see if their math is strong enough to find the answers they want. This is where a tool called a "Fisher forecast" comes in. Think of it like a weather forecast for a scientific experiment. Instead of predicting rain, it predicts how precisely a future telescope will be able to measure the secrets of the universe. It's a way to test the blueprint of the experiment to make sure the final result won't be a blurry mess.

This paper introduces a new, high-tech tool called Augur, built by the Dark Energy Science Collaboration (DESC) for the upcoming Vera C. Rubin Observatory. The Rubin Observatory is about to start a ten-year survey of the entire southern sky, and the scientists need to be absolutely sure their math is solid before they begin. The authors of this paper didn't just build Augur; they put it through the wringer. They tested it against other established computer codes and even compared its predictions to a "gold standard" method called nested sampling, which is like running a full, slow-motion simulation of the universe to see what the data would look like.

The main finding is that Augur works. When the scientists used Augur to predict how well the telescope would measure things like the density of dark matter and the behavior of dark energy, its results matched the other trusted codes almost perfectly. They also discovered a few "traps" to avoid: if you set the computer's calculation steps too small, the results get noisy and unreliable, like trying to measure a mountain's height with a ruler that's too sensitive to dust. They found that a specific step size (about 5% of the range they are testing) is the "sweet spot" for getting stable, trustworthy answers.

The paper also checked how well Augur could predict "biases"—what happens if the scientists make a small mistake in their model. They found that Augur is great at spotting these errors when the mistakes are small, but if the mistake gets too big, the tool's simple math breaks down, just as a straight-line map fails when you try to navigate a winding mountain road.

In short, this paper is a "user manual" and a "safety report" rolled into one. It proves that the new Augur tool is a reliable, fast, and accurate way to plan for the future of cosmology. It doesn't discover a new particle or solve the dark energy mystery itself; instead, it gives the scientists the confidence to say, "Our tools are ready, and when the real data starts coming in, we'll be able to trust what we find." The authors suggest that while Augur is currently the best way to do these quick forecasts, the future might hold even faster tools based on different math, but for now, Augur is the trusted guide for the next decade of cosmic exploration.

Technical Summary: Fisher Forecasting for the DESC with Augur

Problem Statement
The Vera C. Rubin Observatory's Legacy Survey of Space and Time (LSST) is set to deliver unprecedented constraints on cosmological parameters through its ten-year survey. To ensure robust cosmological measurements, the Dark Energy Science Collaboration (DESC) requires computationally inexpensive methods to gauge the performance of proposed analyses and test modeling choices before data collection. While full Markov Chain Monte Carlo (MCMC) or nested sampling analyses provide precise constraints, they are often too computationally intensive for rapid exploration of model spaces. The Fisher Information Matrix formalism offers a faster alternative by estimating parameter constraints via numerical derivatives of the log-likelihood. However, the reliability of Fisher forecasts depends heavily on the assumption that the log-posterior is locally well-approximated by a quadratic function and on the numerical stability of the derivative calculations. Previous LSST DESC work established requirements for systematic effects but lacked a publicly available, unified forecasting library integrated with the DESC software ecosystem.

Methodology
This paper introduces Augur, a Python-based Fisher forecasting tool developed by the LSST DESC. Augur is designed to interface directly with the existing DESC software stack, including:

CCL (Core Cosmology Library): For theoretical predictions of cosmological observables.
TJPCov: For analytic covariance estimation.
firecrown: The official DESC likelihood framework.

The tool computes Fisher matrices by evaluating numerical derivatives of model predictions with respect to cosmological and systematic parameters. The authors validate Augur using a "3×2pt" analysis (combining galaxy clustering, cosmic shear, and galaxy-galaxy lensing) in harmonic space, modeling Year 1 (Y1) and Year 10 (Y10) LSST survey configurations.

Key methodological components include:

Numerical Differentiation: The study investigates various differentiation techniques, including the 5-point stencil method, numdifftools, and derivkit (using adaptive Chebyshev polynomials).
Hyperparameter Stability: The authors systematically test the stability of forecasts against:
- Step Size: The magnitude of parameter perturbations used for derivatives.
- Redshift Distribution ( $n(z)$ ): The sampling resolution (number of knots) of the source and lens galaxy redshift distributions.
- Power Spectrum Models: Comparing linear (Eisenstein-Hu, CAMB) and non-linear (Halofit, HMCode2020) prescriptions.
- Binning: The impact of bin-averaging angular power spectra (top-hat filters).
Validation Benchmarks: Augur's outputs are compared against:
- The legacy DESC Science Requirements Document (SRD) pipeline (using a modified CosmoLike).
- CosmoSIS with firecrown.
- Direct posterior sampling via PolyChord nested sampling.
- Internal consistency checks for Fisher bias estimation.

Key Contributions

Augur Framework: The release of a public, DESC-native forecasting tool that unifies theoretical predictions, likelihood evaluation, and covariance estimation within a single ecosystem.
Stability Diagnostics: The paper establishes specific "working regions" for numerical hyperparameters. It identifies that a normalized step size of approximately 5% of the uniform prior range (e.g., $\Delta s = 0.05$ ) yields stable results, whereas smaller step sizes introduce numerical noise, particularly with non-linear power spectrum models like Halofit.
Model Sensitivity: The study demonstrates that different non-linear power spectrum prescriptions (e.g., Halofit vs. HMCode2020) can lead to significant variations (up to 10% or more) in Figures of Merit (FOMs) even within stable step-size regimes, highlighting the sensitivity of Fisher forecasts to modeling choices.
Bias Estimation: The tool includes functionality to estimate parameter biases arising from model mismatches, with the paper defining the limits of validity for these linear approximations.

Results

Agreement with External Codes: Augur produces results consistent with the CosmoSIS + firecrown pipeline (near-exact agreement in correlation matrices) and the DESC SRD pipeline (DEFOM agreement within 10% for Y1 and 2% for Y10). Discrepancies with the SRD pipeline are attributed to differences in $n(z)$ binning and integration methods.
Validation against Nested Sampling: For stable step sizes ( $\Delta s = 0.05$ ), Augur's Fisher contours closely reproduce the shape, orientation, and area of parameter posteriors obtained via PolyChord nested sampling. The correlation matrices between Fisher and nested sampling results differ by less than 0.05, well within the authors' acceptance threshold of 0.1.
Instability Risks: The study confirms that unstable step sizes (too small) or inadequate $n(z)$ sampling can lead to unreliable FOMs and incorrect degeneracy directions. Specifically, step sizes below 1% of the prior width often result in numerical noise dominating the derivatives.
Bin-Averaging: The inclusion of top-hat bandpower filters smooths variations in FOMs across step sizes, mitigating some numerical instabilities associated with single-mode evaluations.
Fisher Bias Limits: Internal consistency tests show that Fisher bias estimates accurately recover small input parameter shifts (within $\sim 1\sigma$ ). However, for larger shifts (exceeding $\sim 1\sigma$ in $w_0$ or $0.5\sigma$ in $w_a$ ), the linear approximation breaks down, and the bias is not robustly recovered.

Significance and Claims
The authors explicitly state that the primary goal of this work is not to provide updated cosmological forecasts, but rather to introduce and validate a forecasting framework. The paper claims that Augur provides a stable, reliable, and reproducible method for generating Fisher forecasts within the DESC software ecosystem. By documenting best practices for numerical derivatives and modeling choices, the tool allows users to obtain reliable diagnostics for future LSST analyses. The work serves as a methods and validation study, demonstrating that Fisher forecasts can be produced with high fidelity to full likelihood sampling when appropriate numerical stability criteria are met. The authors note that while the current implementation relies on finite-difference derivatives, the stability benchmarks established here will serve as a foundation for future differentiable cosmology frameworks.

More like this