WeSCoS Colloquium #09 “Data-driven prediction and modeling of turbulence based on dynamical systems theory” By Masanobu Inubushi (Associate Professor, Department of Applied Mathematics, Tokyo University of Science)

WeSCoS Colloquium #09 “Data-driven prediction and modeling of turbulence based on dynamical systems theory”
Presenter: Masanobu Inubushi (Associate Professor, Department of Applied Mathematics, Faculty of Science, Tokyo University of Science)
Date and time: 5:00 pm-, July 18, 2024 (Thursday)
Abstract:
Prediction and modeling of turbulence has long been an important issue in fluid mechanics. In recent years, expectations have been rising for data-driven methods that have made rapid progress in this regard, and in this lecture we will introduce an approach based on dynamical systems theory.
In the first half, we will consider low-dimensional modeling of turbulence. At this time, it is important to focus on dependent variables. We investigate from the perspective of data assimilation that small-scale variables become dependent on large-scale variables in turbulent flow, and introduce a framework for characterizing this dependence through stability analysis (using the cross-sectional Lyapunov index) [1]. In the second half, we will discuss prediction and modeling of dynamical systems. Reservoir computing (RC) is a machine learning method that has been actively researched due to its simplicity and remarkable prediction and modeling performance, but there are many unknowns about its mathematical mechanism. In this study, we deepen our mathematical understanding of RC by focusing on generalized synchronization phenomena in dynamical systems, and introduce a new learning method based on this and its performance [2].
[1] M. Inubushi, Y. Saiki, M. U. Kobayashi, and S. Goto, Characterizing Small-Scale Dynamics of Navier-Stokes Turbulence with Transverse Lyapunov Exponents: A Data Assimilation Approach, Phys. Rev. Lett. 131, 254001 (2023 ).
[2] A. Ookubo and M. Inubushi, Reservoir Computing with Generalized Readout based on Generalized Synchronization (arXiv:2405.14885).

Venue: Zoom Pre-registration is required. Please register from here by July 16. We will send the Zoom URL on July 17.
·Language: Japanese

*The WeSCoS colloquium is part of the Moonshot-type R&D project Objective 8.
The colloquium in the core research ‘Control theory of weather-social coupled systems to support social decision-making’ is open to the public.
*WeSCoS stands for Weather-Society Coupling/Control System.