Data-Driven Nonlinear Model Reduction to Spectral Submanifolds via Oblique Projection
Leonardo Bettini, Bálint Kaszás, Bernhard Zybach, Jürg Dual, George Haller · 2025
Read Data-Driven Nonlinear Model Reduction to Spectral Submanifolds via Oblique Projection by Leonardo Bettini, Bálint Kaszás, Bernhard Zybach, Jürg Dual, George Haller (2025) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2503-21895.txt.
Introduction
Five authors led by Leonardo Bettini shrink wild machine motion to a slow curved sheet, even when the frame twists the cast. Such sheets, called spectral submanifolds, host the slow core of a system acts near rest. Far paths are hard to pair with sheet paths when the frame pulls hard off-axis. The true pair path runs down a leaf stack that real data can never map in full.
Their slant cast fakes that leaf stack well from as few as one test run of the rig. A bent beam with harsh twist serves as the proof case in the work. Cast paths land on their true sheet mates at rates past tools miss. Test halls can thus build small true models from thin runs.
Catalog record
| CREATOR | Leonardo Bettini, Bálint Kaszás, Bernhard Zybach, Jürg Dual, George Haller |
|---|---|
| YEAR | 2025 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 47 KB |
| VERSION | 1.0 |
| LICENSE | http://creativecommons.org/licenses/by/4.0/ |
| RIGHTS | CC BY |
| SOURCE | arXiv |
Source: https://arxiv.org/abs/2503-21895
Browse the full library · About the Tower of Babel
Full text
The opening of the work, from the library's hosted text:
Data-Driven Nonlinear Model Reduction to Spectral Submanifolds via Oblique Projection Preprint: AIP/123-QED Leonardo Bettini, Bálint Kaszás, Bernhard Zybach, Jürg Dual, George Haller Abstract (To appear in Chaos ) The dynamics in a primary Spectral Submanifold (SSM) constructed over the slowest modes of a dynamical system provide an ideal reduced-order model for nearby trajectories. Modeling the dynamics of trajectories further away from the primary SSM, however, is difficult if the linear part of the system exhibits strong non-normal behavior. Such non-normality implies that simply projecting trajectories onto SSMs along directions normal to the slow linear modes will not pair those trajectories correctly with their reduced counterparts on the SSMs. In principle, a well-defined nonlinear projection along a stable invariant foliation exists and would exactly match the full dynamics to the SSM-reduced dynamics. This foliation, however, cannot realistically be constructed from practically feasible amounts and distributions of experimental data. Here we develop an oblique projection technique that is able to approximate this foliation efficiently, even from a single experimental trajectory of a significantly non-normal and nonlinear beam. Data-driven reduced-order models seeking to describe, predict and control complex physical systems attract growing interest in various areas of applied science and engineering. Reduction to spectral submanifolds (or SSMs) provides a mathematically justified model reduction approach that captures the nonlinear dynamics on low-dimensional, attracting invariant manifolds in the phase space of the system. Nearby trajectories approach SSMs and quickly synchronize with their internal dynamics, which then serve as accurate reduced models for the longer-term dynamics of the full system. Therefore, establishing a proper correspondence between a trajectory outside an SSM and its target trajectory within the SSM is essential. This task becomes challenging when the linear part of the system is strongly non-normal, necessitating an oblique projection of trajectories onto their targets within the SSM. Here, we develop a data-driven identification of this oblique projection and allows accurate SSM reduction from a few experimentally observed trajectories of highly non-normal and nonlinear systems. I Introduction Reduced-order models have become crucially important in studying high-dimensional nonlinear dynamical systems across various fields of applied science and engineering. They allow us to reduce computational cost while retaining physical interpretability and enabling fast parametric analysis, especially when only experimental data are available. Numerous model reduction methods have been proposed using vastly different approaches (see Benner, Gugercin, and Willcox 2015 ; Rowley and Dawson 2017 ; Taira et al. 2017 ; Brunton, Noack, and Koumoutsakos 2020 ; Touzé, Vizzaccaro, and Thomas 2021 ; Ghadami and Epureanu 2022 for recent reviews). Proper orthogonal decomposition (POD), followed by Galerkin projection is a popular and simple approach to reduce the dimension of nonlinear dynamical systems. It employs a normal projection onto the most energetic modes of the system. However, the main limitation of projecting onto linear subspaces is that those subspaces are not invariant under the nonlinear dynamics ( Ohlberger and Rave 2016 ). This motivates different techniques that project the dynamics onto general invariant manifolds, whose internal dynamics serve as a reduced order model for the system. We employ the theory of spectral submanifolds (SSMs), which allows us to perform accurate model reduction for essentially nonlinear phenomena. Such SSMs rigorously extend the concept of nonlinear normal modes (NNMs) originally introduced by Shaw and Pierre 1993 ; Shaw and Pierre 1994 and Shaw, Pierre, and Pesheck 1999 . They have been mathematically formalized by Cabré, Fontich, and De La Llave 2005 and Haller and Ponsioen 2016 for systems with asymptotically stable stationary states. In particular, the primary SSM is the smoothest continuation of a spectral subspace of the linearized system at the stationary state under the inclusion of nonlinearities. Its existence, uniqueness and persistence in both autonomous and non-autonomous systems depend on certain non-resonance conditions among eigenvalues of the spectrum of the linearization (see Haller and Ponsioen 2016 ). The internal dynamics within the primary SSM tangent to the spectral subspace spanned by the slowest modes is an ideal nonlinear reduced model for the system. Indeed, such an SSM attracts all nearby trajectories, each of which then synchronizes exponentially fast with a particular trajectory within the SSM. Therefore, properly identifying such a target trajectory on the SSM for each off-SSM initial condition is crucial for effective reduced-order modelling. Classic linearization results near hyperbolic fixed points guarantee that off-SSM initial conditions converging the fastest to a given on-SSM initial condition form an (n-d) -dimensional surface, where n is the dimension of the full system and d is the dimension of the SSM (see Haller and Kaundinya 2024 ). This surface is usually called the stable fiber associated with the on-SSM initial condition, which is referred to as the base point of the stable fiber. Stable fibers are known to be as smooth as the dynamical system and provide a smooth foliation of a neighborhood of the SSM. This stable foliation is invariant as fibers are mapped into fibers by the full flow map (see Szalai 2023 for more discussion). If the initial condition lies close to the SSM, the SSM is nearly flat (this is often the case in a delay-embedding setting, see Cenedese et al. 2022a ) and the linear part is close to normal, then projecting such initial condition orthogonally onto the SSM yields an accurate reduced initial condition whose trajectory within the SSM serves as a target for the trajectory of the full system. Such a normal projection approximates the stable fibers of the SSM as planes aligned with the direction of the fast subspace. In our work, orthogonality or normality refers to the mutual orientation of eigenspaces in the phase space. The above SSM-reduction combined with orthogonal projection has proven successful in both equation- and data-driven problems, representing essentially nonlinear features. In the data-driven case, unforced, decaying data is employed to fit the parametrization of the primary SSM and its reduced dynamics. External forcing can then be taken into account to predict the nonlinear response of the system (see Haller and Ponsioen 2016 ; Cenedese et al. 2022a ; Cenedese et al. 2022b ; Axås, Cenedese, and Haller 2023 ; Cenedese et al. 2023 ; Haller and Kaundinya 2024 ), even when non-smoothness is involved ( Bettini, Cenedese, and Haller 2024 ). Recent work by Haller et al. 2023 has revealed the existence of an additional, infinite family of fractional (or secondary) SSMs in C^{\infty} dynamical systems. These are tangent to the same spectral subspace as the primary SSM, but they fill an entire open set of the domain of attraction of the fixed point. Fractional SSMs are of lower smoothness class than the primary SSM, which is indicated by the appearance of non-integer powers in their parametrization. This new class of SSMs arises in modeling transitions between isolated states ( Kaszás and Haller 2024 ) or in experimental problems, as general initial conditions lie on a fractional SSM with probability one. Haller et al. 2023 also extend the theory of SSMs to spectral subspaces with both stable and unstable directions (mixed-mode SSMs). Among other techniques exploiting manifold reduction through normal projection, Lee and Carlberg 2019 ; Champion et al. 2019 ; Fresca, Dede’, and Manzoni 2021 ; Conti et al. 2023 and Romor, Stabile, and Rozza 2023 employ an autoencoder.