Recovering Missing Degrees of Freedom in Simulation-Trained Field-Level Generative Models
astro-ph.CO, cs.LG
Submitted: 2026-11-02
Updated: 2026-11-12
Comments: 15 pages, 9 figures. Expanded manuscript prepared for submission to Phys. Rev. D
License: http://creativecommons.org/licenses/by/4.0/
The gist: Simulation-trained generative models are emerging as a powerful framework for field-level inference in cosmology…
Terminology
Abstract
Simulation-trained generative models are emerging as a powerful framework for field-level inference in cosmology, particularly in strongly nonlinear regimes where analytical approaches become intractable. However, the finite size of training datasets often leads to variance shrinkage, indicating that part of the realization-level variability of the underlying physical process is not captured by the learned generative dynamics. In this paper, we interpret variance shrinkage as a diagnostic of missing effective degrees of freedom and introduce physics-informed latent modifications to the generative dynamics of a continuous normalizing flow. We show that these latent corrections recover a substantial fraction of the missing variability without requiring additional training simulations. Using weak-lensing field-level inference as a representative cosmological application, we further demonstrate through posterior analysis that the learned latent directions are actively exploited during inference and behave as effective nuisance parameters that absorb part of the mismatch between the training and target distributions. As a result, the model exhibits improved robustness to out-of-distribution shifts in field-level inference. Our work establishes variance shrinkage as a practical diagnostic for identifying missing degrees of freedom in simulation-trained generative models and provides a framework for recovering them through physics-informed modifications of generative dynamics.
Sources
- Bayesian physical reconstruction of initial conditions from large scale structure surveys
- Unbiased Cosmology Inference from Biased Tracers using the EFT Likelihood
- Field-level inference in cosmology
- Neural style transfer of weak lensing mass maps
- Replicating weak-lensing summary-statistic covariances with normalizing flows
- Learning the Universe: Posterior Reliability of Neural Generative Models in High-Dimensional Field-Level Inference of Cosmic Initial Conditions
- Machine Learning Techniques for Astrophysics and Cosmology: Simulation-Based Inference
- Weak Lensing Tomographic Redshift Distribution Inference for the Hyper Suprime-Cam Subaru Strategic Program three-year shape catalogue
- Fourier Neural Operator for Parametric Partial Differential Equations
- Flow Matching for Generative Modeling
- $\kappa$TNG: Effect of Baryonic Processes on Weak Lensing with IllustrisTNG Simulations
- Simulations of Wide-Field Weak Lensing Surveys I: Basic Statistics and Non-Gaussian Effects
- Power Spectrum Super-Sample Covariance
- Hyper Suprime-Cam Year 3 Results: Cosmology from Cosmic Shear Two-point Correlation Functions
- BlackJAX: Composable Bayesian inference in JAX
- Weak Gravitational Lensing
- First results from the IllustrisTNG simulations: the stellar mass content of groups and clusters of galaxies
- FiLM: Visual Reasoning with a General Conditioning Layer
- Gaussian Error Linear Units (GELUs)
- Super-sample covariance of the power spectrum, bispectrum, halos, voids, and their cross covariances
Related papers
- 21cmEMUv3: a hybrid diffusion-LSTM emulator of 21cmFAST summary observables
- Connecting field-level and n-point function measurements of galaxy bias
- The Impact of Cosmic Variance and Satellites on JWST Clustering Measurements at Redshift around 6
- Universality of Primordial Anisotropies in Gravitational Wave Background
- KiDS-1000 cosmic shear reanalysis using MetaCalibration
- 2Flow2SN: Dusty Supernova Cosmology with Photometric Redshifts and Contamination using Neural Simulation-Based Inference