TL;DR
This paper introduces a novel multiscale model reduction technique for simulating fluid flow and transport in thin, complex domains with reactive boundaries. By developing specialized multiscale basis functions, the method significantly reduces computational cost while maintaining high accuracy, making it suitable for applications like blood flow simulation in vessels.
What Problem Does This Paper Address?
Many real-world phenomena, such as blood flow in arteries or fluid transport in fractured rock formations, occur in thin domains with intricate geometries and reactive surfaces. Simulating these phenomena with traditional fine-grid methods leads to computationally prohibitive systems of equations. Existing reduced-order models often rely on pre-defined analytical approaches, which may not capture the full complexity of the physics. This paper addresses the need for a systematic and accurate method to reduce the dimensionality of these problems, enabling efficient simulation without sacrificing precision, especially when dealing with complex boundary conditions.
Methods
The authors propose a multiscale model reduction technique based on the Discontinuous Galerkin Generalized Multiscale Finite Element Method (DG-GMsFEM). The core idea is to construct local multiscale basis functions that capture the essential physics within each coarse grid cell.
For the flow problem, they first generate "snapshot" solutions by solving local problems with various boundary conditions on a fine grid. A dimension reduction is then performed by solving a local spectral problem on these snapshots, selecting eigenvectors corresponding to the smallest eigenvalues as multiscale basis functions. Pressure is approximated using piecewise constant functions on the coarse grid.
For the transport problem, similar multiscale basis functions are constructed for each interface between coarse grid cells. Crucially, they introduce additional basis functions specifically designed to capture the effects of non-homogeneous boundary conditions on the reactive walls. The method considers two types of multiscale spaces: one that combines all flow directions and another that separates them, and similarly for transport, considering different boundary conditions separately.
Datasets
Not discussed in the paper. The authors use three custom-designed computational geometries (Geometry 1, 2, and 3) with varying degrees of complexity and fine/coarse grid resolutions for numerical simulations.
Key Insights and Ideas
- Systematic Dimension Reduction: The paper moves beyond ad-hoc analytical models by providing a systematic, data-driven approach to dimension reduction using multiscale basis functions derived from local spectral problems.
- Handling Reactive Boundaries: A key innovation is the development of specialized basis functions to accurately capture the influence of non-homogeneous reactive boundary conditions, which are critical in many transport problems.
- Flexibility in Basis Construction: The authors explore different strategies for constructing multiscale basis functions (e.g., separating flow directions) to optimize performance for specific problem types.
- Accuracy vs. Computational Cost: The results demonstrate a significant reduction in the number of degrees of freedom (up to 99.7% for velocity) with minimal loss in accuracy, highlighting the efficiency of the proposed method.
- Unstructured Grids: The method's effectiveness is shown to extend to unstructured coarse grids, making it applicable to more complex and realistic geometries.
Key Results
The proposed multiscale method achieves significant computational savings while maintaining high accuracy. For instance, in Test 1 (different geometries), the number of degrees of freedom for velocity was reduced by over 99% (e.g., from 121,450 to 410 for Geometry 1), with relative errors in velocity typically below 4% for the final time step. Concentration errors were also significantly reduced, often below 2% for well-chosen basis functions. The study also showed that for 3D problems, Type 2 basis functions (separating flow directions) generally yield better results than Type 1. For transport problems, incorporating velocity information into basis construction generally improved accuracy, especially for Neumann boundary conditions.
Limitations and Future Work
The authors acknowledge that the choice of the number of multiscale basis functions (Mu and Mc) can influence accuracy, and suggest that a sufficient number is needed for good results. They also note that for homogeneous boundary conditions, Type 2 basis functions can be simplified, reducing the system size. Future work could involve further optimization of the basis construction strategies, exploring adaptive methods for selecting basis functions, and extending the approach to more complex physics like turbulence or multiphase flow. The paper also implicitly suggests that a posteriori error estimation could be used to further refine the dimension reduction.
Summary updated