Discontinuous Galerkin Model Order Reduction of Geometrically Parametrized Stokes Equation

Nirav Vasant Shah, Martin W. Hess, Gianluigi Rozza

TL;DR

This paper introduces a novel approach to accelerate simulations of fluid flow governed by the Stokes equation when the geometry of the domain changes. By combining the Discontinuous Galerkin Method (DGM) with geometric parametrization and Proper Orthogonal Decomposition (POD), they achieve significant speedups in computations while maintaining accuracy, making it easier to explore a wide range of geometries.

What Problem Does This Paper Address?

Simulating fluid dynamics, particularly for problems like Stokes flow, can be computationally very expensive, especially when the geometry of the domain is not fixed but changes. This is common in applications like shape optimization or when studying the effects of varying physical boundaries. Traditional methods require re-meshing and re-solving the entire problem for each new geometry, which is prohibitively slow for exploring many different configurations. The challenge is to develop a method that can efficiently predict the flow behavior for a range of geometries without performing a full, costly simulation each time.

Methods

The authors propose a framework that integrates several key techniques:

  • Discontinuous Galerkin Method (DGM): This numerical method is used for solving the full-order model (FOM). DGM is chosen for its flexibility, particularly its ability to handle complex geometries and its polynomial approximation capabilities, which are beneficial for accuracy.
  • Geometric Parametrization: The geometry of the domain is described by a set of parameters. The paper focuses on an "affine parametrization," meaning the transformation from a fixed reference domain to the actual, parametrized domain is linear with respect to these parameters. This affine property is crucial for the subsequent model order reduction step.
  • Proper Orthogonal Decomposition (POD): This is a dimensionality reduction technique. The authors generate "snapshots" of the flow solution (velocity and pressure) for various geometries. POD then identifies the most important "modes" or basis functions that capture the dominant behavior of these solutions. This creates a low-dimensional "reduced basis" space.
  • Galerkin Projection: The original, high-dimensional DGM system is projected onto this low-dimensional reduced basis space. This results in a much smaller system of equations that can be solved much faster.
  • Offline-Online Decomposition: The computational cost is split into an "offline" phase (generating snapshots, computing the reduced basis) and an "online" phase (solving the reduced system for a new parameter/geometry). The affine parametrization allows for efficient evaluation of the system matrices and vectors in the online phase.

Datasets

Not discussed in the paper. The authors use a specific test problem with a defined obstacle geometry and parameter ranges for their numerical experiments.

Key Insights and Ideas

  • Bridging Geometry and Dynamics: The core innovation is effectively linking geometric variations to the underlying fluid dynamics solution through an affine parametrization. This allows the reduced order model to "learn" how changes in shape affect the flow.
  • DGM's Role in Parametrization: DGM's ability to handle discontinuous approximations and its formulation are well-suited for the geometric transformations required by the affine expansion.
  • Speedup through Dimensionality Reduction: POD is shown to be highly effective in capturing the essential solution variations across different geometries, leading to substantial computational speedups.
  • Affinity is Key: The affine nature of the geometric transformation is critical for the efficiency of the offline-online decomposition, enabling rapid evaluation of the reduced system for new parameters.

Key Results

The paper demonstrates the effectiveness of their method on an obstacle test problem.

  • Significant Speedup: The online phase of the reduced system was solved in 0.009 seconds, compared to 6.74 seconds for the full-order model. The overall speedup achieved was approximately 20.6.
  • Accuracy Maintained: The method maintains reliability, with error analysis showing that the reduced basis approach provides solutions close to the full-order model, especially with a sufficiently sized reduced basis.
  • Eigenvalue Decay: Figures illustrate a rapid decay in eigenvalues, indicating that a low-dimensional basis can effectively capture the solution space.

Limitations and Future Work

  • Limitations: The paper acknowledges that the convergence analysis is not discussed in detail. The method's performance is demonstrated on a specific Stokes flow problem; its applicability to more complex flow regimes (e.g., Navier-Stokes with turbulence) might require further investigation. The affine parametrization assumption, while powerful, might not hold for all types of geometric variations.
  • Future Work: The authors suggest that this work contributes to exploring further potentials in geometric parametrization and reduced basis approaches for DGM. Future research could focus on extending this to more complex fluid dynamics problems, investigating non-affine parametrizations, and developing more robust error estimation techniques for parametrized DGM.

Summary updated

Abstract

The present work focuses on the geometric parametrization and the reduced order modeling of the Stokes equation. We discuss the concept of a parametrized geometry and its application within a reduced order modeling technique. The full order model is based on the discontinuous Galerkin method with an interior penalty formulation. We introduce the broken Sobolev spaces as well as the weak formulation required for an affine parameter dependency. The operators are transformed from a fixed domain to a parameter dependent domain using the affine parameter dependency. The proper orthogonal decomposition is used to obtain the basis of functions of the reduced order model. By using the Galerkin projection the linear system is projected onto the reduced space. During this process, the offline-online decomposition is used to separate parameter dependent operations from parameter independent operations. Finally this technique is applied to an obstacle test problem.The numerical outcomes presented include experimental error analysis, eigenvalue decay and measurement of online simulation time. Keywords: Discontinuous Galerkin method, Stokes flow, Geometric parametrization, Proper orthogonal decomposition
BibTeX
@inproceedings{shah2019discontinuous,
  author = {Nirav Vasant Shah and Martin W. Hess and Gianluigi Rozza},
  title = {{Discontinuous Galerkin Model Order Reduction of Geometrically Parametrized Stokes Equation}},
  booktitle = {Enumath},
  year = {2019},
  doi = {10.1007/978-3-030-55874-1\_54},
}
Figure