Abstract:The domain decomposition (DD) nonlinear-manifold reduced-order model (NM-ROM) represents a computationally efficient method for integrating underlying physics principles into a neural network-based, data-driven approach. Compared to linear subspace methods, NM-ROMs offer superior expressivity and enhanced reconstruction capabilities, while DD enables cost-effective, parallel training of autoencoders by partitioning the domain into algebraic subdomains. In this work, we investigate the scalability of this approach by implementing a "bottom-up" strategy: training NM-ROMs on smaller domains and subsequently deploying them on larger, composable ones. The application of this method to the two-dimensional time-dependent Burgers' equation shows that extrapolating from smaller to larger domains is both stable and effective. This approach achieves an accuracy of 1% in relative error and provides a remarkable speedup of nearly 700 times.

What problem does this paper attempt to address?

The main problem that this paper attempts to solve is to improve the computational efficiency in the simulation of large - scale nonlinear dynamical systems while maintaining high precision. Specifically, the authors focus on how to construct an efficient and scalable model - order - reduction framework by combining the nonlinear manifold reduced - order model (NM - ROM) and the domain decomposition (DD) method. ### Main Problems and Solutions 1. **Trade - off between Computational Efficiency and Precision** - **Problem**: Traditional full - order models (FOM) require a large amount of computational resources and time when dealing with large - scale, parameterized, non - linear systems, which makes complex tasks such as design optimization and uncertainty quantification impractical. - **Solution**: By introducing reduced - order models (ROM), especially the nonlinear manifold reduced - order model (NM - ROM), the amount of computation can be significantly reduced while maintaining relatively high precision. NM - ROM has better expressive and reconstructive capabilities compared to linear subspace methods. 2. **Scalability and Parallel Training** - **Problem**: For large - scale problems, directly training NM - ROM can be very time - consuming and computationally expensive. - **Solution**: Adopt a "bottom - up" strategy, that is, first train NM - ROM on smaller sub - domains and then deploy it to larger, combinable domains. This method not only improves scalability but also allows parallel training of auto - encoders, thereby reducing costs. 3. **Dealing with Problems with Slowly Decaying Kolmogorov n - width** - **Problem**: The Kolmogorov n - width of some physical problems (such as convection - dominated problems and problems with sharp gradients) decays slowly, making it difficult for traditional linear subspace reduced - order models to handle effectively. - **Solution**: By approximating FOM on a low - dimensional nonlinear manifold, NM - ROM can better handle these problems. In addition, combining the domain decomposition method can further improve the applicability and performance of the model. ### Specific Applications In the paper, the two - dimensional time - varying Burgers equation is taken as an example to demonstrate the effectiveness and superiority of this method. The experimental results show that extrapolating from a smaller domain to a larger domain is stable and effective, with a relative error of about 1% and a speed - up of nearly 700 times. ### Summary In summary, this paper aims to solve the computational efficiency and precision problems in the simulation of large - scale nonlinear dynamical systems by combining the nonlinear manifold reduced - order model and the domain decomposition method, and proposes a "bottom - up" training strategy to improve the scalability and parallelization capabilities of the model.

Scalable nonlinear manifold reduced order model for dynamical systems

Nonlinear-manifold reduced order models with domain decomposition

A fast and accurate domain-decomposition nonlinear manifold reduced order model

Model reduction of dynamical systems on nonlinear manifolds using deep convolutional autoencoders

A fast and accurate physics-informed neural network reduced order model with shallow masked autoencoder

Non-linear Manifold Reduced-Order Models with Convolutional Autoencoders and Reduced Over-Collocation Method

Reduced order modeling for flow and transport problems with Barlow Twins self-supervised learning

SNF-ROM: Projection-based nonlinear reduced order modeling with smooth neural fields

Learning physics-based reduced-order models from data using nonlinear manifolds

Non-intrusive model reduction of large-scale, nonlinear dynamical systems using deep learning

A comprehensive deep learning-based approach to reduced order modeling of nonlinear time-dependent parametrized PDEs

Nonlinear Balanced Truncation: Part 2 -- Model Reduction on Manifolds

Nonlinear model order reduction for problems with microstructure using mesh informed neural networks

Domain Decomposition Strategy for Combining Nonlinear and Linear Reduced-Order Models

Non-intrusive model reduction of advection-dominated hyperbolic problems using neural network shift augmented manifold transformations

A manifold learning approach to nonlinear model order reduction of quasi-static problems in solid mechanics

Hyper-Reduced Autoencoders for Efficient and Accurate Nonlinear Model Reductions

Explicable hyper-reduced order models on nonlinearly approximated solution manifolds of compressible and incompressible Navier-Stokes equations

Neural downscaling for complex systems: from large-scale to small-scale by neural operator

Neural Manifold Operators for Learning the Evolution of Physical Dynamics

A Tailored Convolutional Neural Network for Nonlinear Manifold Learning of Computational Physics Data using Unstructured Spatial Discretizations