Abstract:We present a novel physics-constrained polynomial chaos expansion as a surrogate modeling method capable of performing both scientific machine learning (SciML) and uncertainty quantification (UQ) tasks. The proposed method possesses a unique capability: it seamlessly integrates SciML into UQ and vice versa, which allows it to quantify the uncertainties in SciML tasks effectively and leverage SciML for improved uncertainty assessment during UQ-related tasks. The proposed surrogate model can effectively incorporate a variety of physical constraints, such as governing partial differential equations (PDEs) with associated initial and boundary conditions constraints, inequality-type constraints (e.g., monotonicity, convexity, non-negativity, among others), and additional a priori information in the training process to supplement limited data. This ensures physically realistic predictions and significantly reduces the need for expensive computational model evaluations to train the surrogate model. Furthermore, the proposed method has a built-in uncertainty quantification (UQ) feature to efficiently estimate output uncertainties. To demonstrate the effectiveness of the proposed method, we apply it to a diverse set of problems, including linear/non-linear PDEs with deterministic and stochastic parameters, data-driven surrogate modeling of a complex physical system, and UQ of a stochastic system with parameters modeled as random fields.

What problem does this paper attempt to address?

The main problem that this paper attempts to solve is to develop a new Physics - Constrained Polynomial Chaos Expansion (PC2) method as a surrogate modeling technique to perform scientific machine learning (SciML) and uncertainty quantification (UQ) tasks simultaneously. Specifically, this method aims to: 1. **Seamlessly integrate SciML and UQ**: By incorporating physical constraints (such as partial differential equations and their initial and boundary conditions, inequality constraints, etc.) into the polynomial chaos expansion, the PC2 method can effectively quantify uncertainty in SciML tasks and utilize SciML to improve the accuracy of uncertainty assessment in UQ tasks. 2. **Ensure physical realism**: By introducing various physical constraints and prior information during the training process, ensure that the model prediction results conform to physical laws, thereby providing more reliable predictions. 3. **Reduce computational cost**: By using virtual collocation points to impose constraints instead of performing a large number of evaluations on expensive computational models, significantly reduce the computational resources required to train the surrogate model. 4. **Handle high - dimensional problems**: By combining sparse regression algorithms (such as Least Angle Regression (LAR)), reduce the number of polynomial basis functions, making the PC2 method suitable for high - dimensional problems. 5. **Wide applicability**: This method can be applied to various types of scientific and engineering problems, including linear/non - linear partial differential equations, data - driven surrogate modeling of stochastic parameters, and uncertainty quantification of stochastic systems. ### Specific Objectives - **Solve deterministic and stochastic partial differential equations**: By incorporating PDE constraints and their initial and boundary conditions into the PC2 method, solve deterministic and stochastic partial differential equations. - **Data - driven surrogate modeling**: Perform data - driven surrogate modeling of complex physical systems when physical constraints are known. - **Uncertainty quantification**: Quantify the uncertainty of parameters (such as random fields) in stochastic systems and provide reliable output uncertainty estimates. ### Method Overview - **Polynomial Chaos Expansion (PCE)**: PCE is a commonly used uncertainty quantification method that approximates the response of a stochastic system through polynomial expansion. - **Introduction of physical constraints**: By introducing physical constraints such as PDEs, initial and boundary conditions, and inequality constraints in the PCE framework, ensure that the model prediction conforms to physical laws. - **Virtual collocation points**: Impose constraints at discrete positions in the input domain. These positions are called virtual collocation points, and no evaluation of expensive computational models is required. - **Solution of optimization problems**: By transforming constraint conditions into soft constraints, transform the optimization problem into an unconstrained optimization problem, and use an adaptive weighting scheme to balance the influence of different loss terms. ### Application Examples - **Linear/non - linear PDEs**: Solve linear/non - linear partial differential equations with deterministic and stochastic parameters. - **Data - driven surrogate modeling**: Perform data - driven surrogate modeling of complex physical systems, for example, based on experimental or low - fidelity simulation data. - **Uncertainty quantification of stochastic systems**: Quantify the uncertainty of parameters (such as random fields) in stochastic systems. Through these methods and applications, this paper demonstrates the effectiveness and superiority of the PC2 method in dealing with complex scientific and engineering problems.

Physics-constrained polynomial chaos expansion for scientific machine learning and uncertainty quantification

Fast Uncertainty Quantification by Sparse Data Learning from Multiphysics Systems

Physics-Informed Polynomial Chaos Expansions

Surrogate Modeling of High-Dimensional Problems Via Data-Driven Polynomial Chaos Expansions and Sparse Partial Least Square

Polynomial Chaos Expansions on Principal Geodesic Grassmannian Submanifolds for Surrogate Modeling and Uncertainty Quantification

3D multi-physics uncertainty quantification using physics-based machine learning

Uncertainty Quantification of Surrogate Models using Conformal Prediction

Stochastic polynomial chaos expansions to emulate stochastic simulators

A Generalized Multi-Fidelity Simulation Method Using Sparse Polynomial Chaos Expansion

Multivariate sensitivity-adaptive polynomial chaos expansion for high-dimensional surrogate modeling and uncertainty quantification

A New Surrogate Modeling Method Combining Polynomial Chaos Expansion And Gaussian Kernel In A Sparse Bayesian Learning Framework

Uncertainty Quantification For Turbulent Flows with Machine Learning

Polynomial Chaos Surrogate Construction for Random Fields with Parametric Uncertainty

Sparse approximation of data-driven Polynomial Chaos expansions: an induced sampling approach

NeuralUQ: A Comprehensive Library for Uncertainty Quantification in Neural Differential Equations and Operators

Uncertainty quantification for noisy inputs-outputs in physics-informed neural networks and neural operators

Uncertainty Quantification Using Polynomial Chaos Expansion with Points of Monomial Cubature Rules

Leveraging viscous Hamilton-Jacobi PDEs for uncertainty quantification in scientific machine learning

Robust and scalable uncertainty estimation with conformal prediction for machine-learned interatomic potentials

Leveraging Viscous Hamilton–Jacobi PDEs for Uncertainty Quantification in Scientific Machine Learning