Saved in:
Bibliographic Details
Main Authors: Gao, Hang, Kaltenbach, Sebastian, Koumoutsakos, Petros
Format: Recurso digital
Language:English
Published: Zenodo 2025
Online Access:https://doi.org/10.5281/zenodo.14913931
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1866902108952854528
author Gao, Hang
Kaltenbach, Sebastian
Koumoutsakos, Petros
author_facet Gao, Hang
Kaltenbach, Sebastian
Koumoutsakos, Petros
contents <p>We introduce generative models for accelerating simulations of complex systems through learning and evolving their effective dynamics. In the proposed Generative Learning of Effective Dynamics (G-LED), instancesof high dimensional data are down sampled to a lower dimensional manifold that is evolved through an autoregressive attention mechanism. In turn, Bayesian diffusion models, that map this low-dimensional manifold onto its corresponding high-dimensional space, capture the statistics of the system dynamics. We demonstrate the capabilities and drawbacks of G-LED in simulations of several benchmark systems, including the Kuramoto–Sivashinsky (KS) equation, two-dimensional high Reynolds number flow over a backward-facing step, and simulations of three-dimensional turbulent channel flow. The results demonstrate that generative learning offers new frontiers for the accurate forecasting of the statistical properties of complex systems at a reduced computational cost.</p>
format Recurso digital
id zenodo_https___doi_org_10_5281_zenodo_14913931
institution Zenodo
language eng
publishDate 2025
publisher Zenodo
record_format zenodo
spellingShingle Generative Learning for Forecasting the Dynamics of Complex Systems
Gao, Hang
Kaltenbach, Sebastian
Koumoutsakos, Petros
<p>We introduce generative models for accelerating simulations of complex systems through learning and evolving their effective dynamics. In the proposed Generative Learning of Effective Dynamics (G-LED), instancesof high dimensional data are down sampled to a lower dimensional manifold that is evolved through an autoregressive attention mechanism. In turn, Bayesian diffusion models, that map this low-dimensional manifold onto its corresponding high-dimensional space, capture the statistics of the system dynamics. We demonstrate the capabilities and drawbacks of G-LED in simulations of several benchmark systems, including the Kuramoto–Sivashinsky (KS) equation, two-dimensional high Reynolds number flow over a backward-facing step, and simulations of three-dimensional turbulent channel flow. The results demonstrate that generative learning offers new frontiers for the accurate forecasting of the statistical properties of complex systems at a reduced computational cost.</p>
title Generative Learning for Forecasting the Dynamics of Complex Systems
url https://doi.org/10.5281/zenodo.14913931