Abstract:This work formulates a new approach to reduced modeling of parameterized, time-dependent partial differential equations (PDEs). The method employs Operator Inference, a scientific machine learning framework combining data-driven learning and physics-based modeling. The parametric structure of the governing equations is embedded directly into the reduced-order model, and parameterized reduced-order operators are learned via a data-driven linear regression problem. The result is a reduced-order model that can be solved rapidly to map parameter values to approximate PDE solutions. Such parameterized reduced-order models may be used as physics-based surrogates for uncertainty quantification and inverse problems that require many forward solves of parametric PDEs. Numerical issues such as well-posedness and the need for appropriate regularization in the learning problem are considered, and an algorithm for hyperparameter selection is presented. The method is illustrated for a parametric heat equation and demonstrated for the FitzHugh-Nagumo neuron model.

What problem does this paper attempt to address?

The problem that this paper attempts to solve is to develop a new method for the reduced - order modeling of time - dependent partial differential equations (PDEs). Specifically, the authors propose a method based on Operator Inference, which is a scientific machine - learning framework that combines data - driven learning and physical modeling. The main objectives of this method are as follows: 1. **Embed parameter structure**: Embed the parameter structure of the governing equation directly into the reduced - order model, and learn the parameterized reduced - order operator through a data - driven linear regression problem. This enables the rapid solution of the reduced - order model, thereby mapping parameter values to approximate PDE solutions. 2. **Non - invasive model reduction**: This method is non - invasive, that is, it does not require direct access to the underlying code that generates high - fidelity simulation data, but mainly learns the reduced - order model from simulation data and/or outputs. This method combines data - driven learning and physical modeling, and is both flexible and robust. 3. **Solve numerical problems**: Consider numerical problems, such as well - posedness and the need for appropriate regularization in learning problems, and propose a hyper - parameter selection algorithm. 4. **Application examples**: This method is illustrated and verified in the parameterized heat conduction equation and the FitzHugh–Nagumo neuron model. ### Main contributions 1. **General framework**: Propose an OpInf framework applicable to parameterized PDE systems, which deals with affine parameterized problems in a time - continuous setting. 2. **Well - posedness analysis**: Analyze relevant well - posedness conditions, which can be used to guide the selection of parameter samples for training data. 3. **Regularization strategy**: Propose a flexible regularization strategy that can be adjusted according to specific PDE operators to obtain stable and accurate reduced - order models. ### Method overview - **Projection - based reduced - order model**: For parameterized PDE systems, construct a reduced - order model through the projection method, retaining the affine parameterized polynomial structure of the system. - **Data - driven optimization problem**: Learn the reduced - order operator matrix by solving a data - driven least - squares regression problem, requiring only solution samples of parameter values, a finite number of basis functions, and knowledge of the governing equation and parameter structure. - **Well - posedness conditions**: Establish necessary and sufficient conditions to ensure the well - posedness of the problem, guiding the selection of parameter samples. ### Application examples - **Heat conduction equation**: Apply this method in the parameterized heat conduction equation, showing how to accurately capture system dynamics by learning the reduced - order model. - **FitzHugh–Nagumo neuron model**: Further verify the effectiveness of this method in the FitzHugh–Nagumo neuron model. In conclusion, this paper proposes a new non - invasive parameterized reduced - order model method, aiming to efficiently solve the modeling problem of parameterized PDE systems, especially suitable for uncertainty quantification and inverse problems that require multiple forward solutions.

Non-intrusive reduced-order models for parametric partial differential equations via data-driven operator inference

Reduced-order modeling for parameterized PDEs via implicit neural representations

Energy-Preserving Reduced Operator Inference for Efficient Design and Control

Operator inference for non-intrusive model reduction of systems with non-polynomial nonlinear terms

Learning Nonlinear Reduced Models from Data with Operator Inference

Gradient Preserving Operator Inference: Data-Driven Reduced-Order Models for Equations with Gradient Structure

Data-Driven Deep Learning of Partial Differential Equations in Modal Space

Non-Intrusive Reduced Models based on Operator Inference for Chaotic Systems

Non-intrusive data-driven reduced-order modeling for time-dependent parametrized problems

Integration of physics-informed operator learning and finite element method for parametric learning of partial differential equations

A non-intrusive bi-fidelity reduced basis method for time-independent problems

Neural Operator: Is data all you need to model the world? An insight into the impact of Physics Informed Machine Learning

A non-intrusive neural network model order reduction algorithm for parameterized parabolic PDEs

Bayesian learning with Gaussian processes for low-dimensional representations of time-dependent nonlinear systems

Operator inference with roll outs for learning reduced models from scarce and low-quality data

Data-driven reduced-order models via regularized operator inference for a single-injector combustion process

Probabilistic error estimation for non-intrusive reduced models learned from data of systems governed by linear parabolic partial differential equations

Finite Operator Learning: Bridging Neural Operators and Numerical Methods for Efficient Parametric Solution and Optimization of PDEs

A Deep Learning approach to Reduced Order Modelling of Parameter Dependent Partial Differential Equations

Reduced-order Modelling of Parameter-Dependent, Linear and Nonlinear Dynamic Partial Differential Equation Models.