Differentiable Fuzzy Cosmic-Web Model for Improved Field Inference

This paper introduces a differentiable model for field-level inference of the cosmological large-scale structure, addressing the challenge of developing effective field-level bias models within Bayesian reconstruction methods. The model integrates augmented Lagrangian perturbation theory, nonlinear, nonlocal, and stochastic biasing, providing a comprehensive framework for analyzing galaxy surveys. A key innovation is the use of sigmoid-based gradient operations, which enable a fuzzy and differentiable hierarchical cosmic-web description, making the model well-suited for machine learning frameworks. The model is implemented in JaX, enabling GPU-accelerated computations and scalable evaluation of complex biasing. The authors demonstrate that their approach accurately reproduces the primordial density field and reconstructs bias parameters within a Bayesian framework. This research contributes to the ongoing effort to develop more accurate and efficient methods for analyzing galaxy surveys and understanding the formation of the cosmic large-scale structure. The ability to accurately model tracer bias is crucial for extracting meaningful cosmological information from galaxy surveys, influencing studies of dark matter, dark energy, and galaxy evolution.

_Context: This intelligence report was compiled by the DailyOrbitalWire Strategy Engine. Verified for Art. 50 Compliance._