Abstract:Numerical solvers of partial differential equations (PDEs) have been widely employed for simulating physical systems. However, the computational cost remains a major bottleneck in various scientific and engineering applications, which has motivated the development of reduced-order models (ROMs). Recently, machine-learning-based ROMs have gained significant popularity and are promising for addressing some limitations of traditional ROM methods, especially for advection dominated systems. In this chapter, we focus on a particular framework known as Latent Space Dynamics Identification (LaSDI), which transforms the high-fidelity data, governed by a PDE, to simpler and low-dimensional latent-space data, governed by ordinary differential equations (ODEs). These ODEs can be learned and subsequently interpolated to make ROM predictions. Each building block of LaSDI can be easily modulated depending on the application, which makes the LaSDI framework highly flexible. In particular, we present strategies to enforce the laws of thermodynamics into LaSDI models (tLaSDI), enhance robustness in the presence of noise through the weak form (WLaSDI), select high-fidelity training data efficiently through active learning (gLaSDI, GPLaSDI), and quantify the ROM prediction uncertainty through Gaussian processes (GPLaSDI). We demonstrate the performance of different LaSDI approaches on Burgers equation, a non-linear heat conduction problem, and a plasma physics problem, showing that LaSDI algorithms can achieve relative errors of less than a few percent and up to thousands of times speed-ups.

What problem does this paper attempt to address?

The main problem that this paper attempts to solve is: in high - fidelity simulations, due to the high computational cost, there are huge challenges in the simulation of multi - scale and complex physical phenomena. Specifically, although traditional numerical solution methods (such as the finite - difference method, the finite - volume method, the finite - element method, etc.) can provide accurate results, their computational resource requirements are huge, especially when it comes to scenarios such as uncertainty quantification, inverse problems, design optimization, and optimal control, and this requirement becomes more prominent. To solve this bottleneck problem, researchers have developed Reduced - Order Models (ROMs), aiming to reduce the computational complexity by simplifying the original problem. However, traditional ROM methods have limitations in some cases (for example, in convection - dominated systems). Therefore, this paper focuses on introducing a framework called Latent Space Dynamics Identification (LaSDI), which transforms high - fidelity data into low - dimensional latent - space data described by ordinary differential equations (ODEs). These ODEs can be learned and interpolated for ROM prediction. The main contributions of the LaSDI framework include: 1. **Flexibility**: Each building block can be adjusted according to the application, making the LaSDI framework highly flexible. 2. **Thermodynamic Constraints**: By introducing the laws of thermodynamics (tLaSDI), the accuracy of the model is enhanced. 3. **Noise Resistance**: Through the weak form (WLaSDI), the robustness in a noisy environment is improved. 4. **Efficient Data Selection**: Through active learning (gLaSDI, GPLaSDI), high - fidelity training data is efficiently selected. 5. **Uncertainty Quantification**: Through the Gaussian process (GPLaSDI), the uncertainty of ROM prediction is quantified. Through these improvements, the LaSDI algorithm can achieve a relative error of less than a few percent in multiple examples such as the Burgers equation, the nonlinear heat conduction problem, and the plasma physics problem, and accelerate up to thousands of times. In summary, this paper attempts to solve the computational cost problem in high - fidelity simulations through the LaSDI framework, while improving the applicability and accuracy of ROM in complex physical systems.

A Comprehensive Review of Latent Space Dynamics Identification Algorithms for Intrusive and Non-Intrusive Reduced-Order-Modeling

GPLaSDI: Gaussian Process-based Interpretable Latent Space Dynamics Identification through Deep Autoencoder

Weak-Form Latent Space Dynamics Identification

Data-Driven Autoencoder Numerical Solver with Uncertainty Quantification for Fast Physical Simulations

Physics-informed active learning with simultaneous weak-form latent space dynamics identification

Latent-space time evolution of non-intrusive reduced-order models using Gaussian process emulation

On latent dynamics learning in nonlinear reduced order modeling

Nonintrusive projection-based reduced order modeling using stable learned differential operators

Train Small, Model Big: Scalable Physics Simulators via Reduced Order Modeling and Domain Decomposition

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

Modeling Advection-Dominated Flows with Space-Local Reduced-Order Models

Deep learning-based reduced-order methods for fast transient dynamics

A LSTM-enhanced surrogate model to simulate the dynamics of particle-laden fluid systems

Parametric Taylor series based latent dynamics identification neural networks

tLaSDI: Thermodynamics-informed latent space dynamics identification

Nonlinear dimensionality reduction then and now: AIMs for dissipative PDEs in the ML era

Preserving Lagrangian structure in data-driven reduced-order modeling of large-scale dynamical systems

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

Adjacency-based, Non-intrusive Reduced-order Modeling for Fluid-Structure Interactions

Non-intrusive reduced-order model for time-dependent stochastic partial differential equations utilizing dynamic mode decomposition and polynomial chaos expansion