2026 workshop program
Peaks as Probes: Simulation-Based Inference of Dark Matter Clustering from Weak-Lensing Convergence Maps
The work
Abstract
Weak gravitational lensing offers a direct window onto the distribution of dark matter across the Universe. This work focuses on weak-lensing peak counts, the local maxima in the convergence (κ) field, which trace the most massive dark matter structures and carry valuable non-Gaussian information that two-point statistics are unable to capture. Because the peak-count likelihood is itself non-Gaussian and difficult to describe with analytic models, simulation-based inference (SBI) is applied, training a neural posterior estimator on a suite of convergence maps spanning [parameter grid] to infer the cosmological parameters S₈ and Ωₘ directly from the peaks. Posterior calibration is validated through coverage tests, and robustness is assessed against shape noise and baryonic feedback. The results are further benchmarked against a convolutional network performing field-level inference on the same maps, isolating how much cosmological information the peaks alone retain. Together, these results demonstrate a fast, well-calibrated, and interpretable route to higher-order weak-lensing inference, one directly applicable to the convergence maps of upcoming Stage-IV surveys such as Rubin LSST and Euclid.