Skip to content
DraconiticResearch
Search Sources API
226,647 studies31 sourcesUpdated 32 min ago
Peer-reviewedCC BY 4.0Monthly Notices of the Royal Astronomical Society

Benchmarking AI-evolved cosmological structure formation

  1. 1University of Chicago Department of Physics, , Chicago, IL 60637 ,
  2. 2High Energy Physics Division, Argonne National Laboratory , 9700 South Cass Avenue, Lemont, IL 60439 ,
  3. 3Computational Science Division, Argonne National Laboratory , 9700 South Cass Avenue, Lemont, IL 60439 ,

Abstract

The potential of deep learning-based image-to-image translations has attracted significant attention as a fast, approximate alternative to cosmological simulations, which would be particularly useful in various contexts, including covariance studies, investigations of systematics, and cosmological parameter inference. We model the evolution of structure formation using U-Net, a widely employed convolutional image translation framework with previous applications in cosmology. To study aspects of learning-based cosmological mappings in a controlled, illustrative setting, we use two approaches for generating suitable cosmological matter fields as datasets: a simple analytical prescription provided by the Zel’dovich approximation, and a standard N-body method using Particle-Mesh simulations. We apply established metrics such as matter power spectra and bispectra, topological indicators, and statistical independence of density fields and study their reproducibility and convergence properties. The U-Net approach performs well only for some of these physical metrics, and accuracy is worse at more nonlinear scales, where the dynamic range in density is larger. Motivated by field-level emulator studies utilizing Eulerian-aware loss functions and beyond pixel-wise mean squared errors, we examine a density-weighted objective. In the PM tests ending at z = 4, where the evolution is only mildly nonlinear, using the Eulerian matter density evaluated at each particle position as a direct weight on that particle’s Lagrangian displacement error improves the U-Net results at smaller scales; this result does not establish comparable accuracy in the strongly nonlinear low-redshift regime.

Related

  1. Evolved stars: dust formation and mass loss
    Luis Velilla-Prieto, Wouter Vlemmings, Guillermo Quintana-Lacaci +18
    arXivPreprintCC BY 4.0Full textInstrumentation & methodsarXiv
  2. A Study on the Evolution of NGC6822. I. Investigating the Star Formation History through Evolved Stellar Populations
    Fatemeh Khatamsaz, Mahdi Abdollahi, Atefeh Javadi +4
    arXivPreprintarXiv licenceFull textGalaxiesSolar & stellar physicsarXiv
  3. A Study on the Evolution of NGC 6822. I. Investigating the Star Formation History through Evolved Stellar Populations
    Fatemeh Khatamsaz, Mahdi Abdollahi, Atefeh Javadi +4
    The Astrophysical JournalPeer-reviewedCC BY 4.0Full textSolar & stellar physicsIOP/AASCrossref
  4. Untangling Similarities and Differences in Star Formation Histories Across Cosmological Simulations
    Alicia M. Savelli, Matthew A. Kustec, Joshua S. Speagle +5
    arXivPreprintCC BY 4.0Full textGalaxiesarXiv
  5. Learning the Universe with PRFM-vol: Introducing a new subgrid model for star formation in cosmological simulations
    Jan D. Burger, Volker Springel, Eve C. Ostriker +7
    arXivPreprintCC BY 4.0Full textGalaxiesarXiv
BibTeX
@article{dong2026benchmarking,
  title = {{Benchmarking AI-evolved cosmological structure formation}},
  author = {Xiaofeng Dong and Nesar Ramachandra and Salman Habib and Katrin Heitmann},
  journal = {Monthly Notices of the Royal Astronomical Society},
  year = {2026},
  month = nov,
  doi = {10.1093/mnras/stag2032},
  url = {https://academic.oup.com/mnras/advance-article/doi/10.1093/mnras/stag2032/9055757},
  note = {CC BY 4.0},
}
RIS
TY  - JOUR
TI  - Benchmarking AI-evolved cosmological structure formation
AU  - Xiaofeng Dong
AU  - Nesar Ramachandra
AU  - Salman Habib
AU  - Katrin Heitmann
PY  - 2026
DA  - 2026/11/14
JO  - Monthly Notices of the Royal Astronomical Society
DO  - 10.1093/mnras/stag2032
UR  - https://academic.oup.com/mnras/advance-article/doi/10.1093/mnras/stag2032/9055757
AB  - The potential of deep learning-based image-to-image translations has attracted significant attention as a fast, approximate alternative to cosmological simulations, which would be particularly useful in various contexts, including covariance studies, investigations of systematics, and cosmological parameter inference. We model the evolution of structure formation using U-Net, a widely employed convolutional image translation framework with previous applications in cosmology. To study aspects of learning-based cosmological mappings in a controlled, illustrative setting, we use two approaches for generating suitable cosmological matter fields as datasets: a simple analytical prescription provided by the Zel’dovich approximation, and a standard N-body method using Particle-Mesh simulations. We apply established metrics such as matter power spectra and bispectra, topological indicators, and statistical independence of density fields and study their reproducibility and convergence properties. The U-Net approach performs well only for some of these physical metrics, and accuracy is worse at more nonlinear scales, where the dynamic range in density is larger. Motivated by field-level emulator studies utilizing Eulerian-aware loss functions and beyond pixel-wise mean squared errors, we examine a density-weighted objective. In the PM tests ending at z = 4, where the evolution is only mildly nonlinear, using the Eulerian matter density evaluated at each particle position as a direct weight on that particle’s Lagrangian displacement error improves the U-Net results at smaller scales; this result does not establish comparable accuracy in the strongly nonlinear low-redshift regime.
SN  - 0035-8711
SN  - 1365-2966
ER  -