English

Mean-Field Simulation-Based Inference for Cosmological Initial Conditions

Cosmology and Nongalactic Astrophysics 2024-10-22 v1 Instrumentation and Methods for Astrophysics Machine Learning

Abstract

Reconstructing cosmological initial conditions (ICs) from late-time observations is a difficult task, which relies on the use of computationally expensive simulators alongside sophisticated statistical methods to navigate multi-million dimensional parameter spaces. We present a simple method for Bayesian field reconstruction based on modeling the posterior distribution of the initial matter density field to be diagonal Gaussian in Fourier space, with its covariance and the mean estimator being the trainable parts of the algorithm. Training and sampling are extremely fast (training: 1h\sim 1 \, \mathrm{h} on a GPU, sampling: 3s\lesssim 3 \, \mathrm{s} for 1000 samples at resolution 1283128^3), and our method supports industry-standard (non-differentiable) NN-body simulators. We verify the fidelity of the obtained IC samples in terms of summary statistics.

Keywords

Cite

@article{arxiv.2410.15808,
  title  = {Mean-Field Simulation-Based Inference for Cosmological Initial Conditions},
  author = {Oleg Savchenko and Florian List and Guillermo Franco Abellán and Noemi Anau Montel and Christoph Weniger},
  journal= {arXiv preprint arXiv:2410.15808},
  year   = {2024}
}

Comments

Accepted for the NeurIPS 2024 workshop Machine Learning and the Physical Sciences; 5 + 4 pages, 3 figures

R2 v1 2026-06-28T19:29:22.613Z