Solving incompressible Navier-Stokes equations on irregular domains and quadtrees by monolithic approach

Hyuntae Cho, Yesom Park, Myungjoo Kang

TL;DR

This paper introduces a second-order monolithic method for solving incompressible Navier-Stokes equations on complex, irregular domains using adaptive quadtree and octree grids. By solving for velocity and pressure simultaneously in a single linear system (the "monolithic" approach) and utilizing a specific Arakawa B-grid layout, the authors achieve high-order accuracy and robust handling of moving boundaries that typically baffle standard projection-based solvers.

What Problem Does This Paper Address?

Simulating fluid flow around complex shapes—like a ship's hull or a turbine blade—usually requires one of two things: a "body-fitted" mesh that bends to match the shape, or a simple Cartesian (rectangular) grid that uses mathematical tricks to "immerse" the shape within it. While Cartesian grids are easier to generate and refine, they often struggle with accuracy near the boundaries.

The industry standard for these simulations is the projection method, which breaks the calculation into steps (first guess the velocity, then correct it to ensure the fluid doesn't compress). However, on irregular or moving domains, the boundary conditions for these intermediate steps become mathematically "coupled" and incredibly difficult to define accurately. This often leads to a drop in numerical precision.

Methods

The authors depart from the standard "staggered" MAC grid used in most fluid solvers and instead adopt the Arakawa B-grid layout. In this setup, velocity variables live at the cell corners (vertices), while pressure variables sit at the cell centers.

Key technical components include:

  • Monolithic Solver: Instead of the multi-step projection approach, they solve the momentum and incompressibility equations simultaneously in one large "saddle-point" linear system. This eliminates the need for artificial boundary conditions for intermediate velocities.
  • Quadtree/Octree Grids: To save computational power, the grid is only refined where it matters (near boundaries or high-vortex areas) using tree-based data structures.
  • Level Set Method: The irregular boundary is tracked using a level set function, which represents the shape as the zero-contour of a higher-dimensional surface.
  • Stabilization: B-grids are notorious for "checkerboard instability" (spurious oscillations in pressure). The authors implement a global stabilizer to smooth out these artifacts without sacrificing accuracy.
  • Moving Least Squares (MLS): For points near the irregular boundary, they use MLS interpolation to maintain second-order accuracy even when the grid doesn't perfectly align with the shape.

Datasets

As a theoretical and numerical study, the paper does not use external "big data" sets. Instead, it evaluates the algorithm against several classic computational fluid dynamics (CFD) benchmarks:

  • Analytic Single-Vortex: A mathematical "ground truth" used to measure exact error rates.
  • Lid-Driven Cavity Flow: A standard test where fluid is pushed inside a square box.
  • Irregular Domains: Custom shapes including circles, ellipses, and a complex "flower" shape.
  • Moving Boundaries: A rotating flower-shaped object to test the solver's ability to handle time-dependent geometry.
  • Two-Phase Flow: A "rising bubble" simulation to demonstrate the method's extensibility to multi-fluid problems.

Key Insights and Ideas

  • The B-Grid Advantage: While MAC grids are popular, they require wide, complex stencils on quadtrees. The Arakawa B-grid allows for compact stencils, making the math much cleaner on non-uniform, adaptive meshes.
  • Simplicity through Complexity: While a monolithic system results in a larger, harder-to-solve matrix than projection methods, it drastically simplifies the logic for boundary conditions. This trade-off is worth it for irregular domains where projection methods lose their "second-order" accuracy.
  • Stabilization is the Secret Sauce: The paper demonstrates that without their specific pressure stabilizer, the pressure convergence rate collapses in the $L^2$ norm. With it, the system remains robust even at high Reynolds numbers (high-speed/low-viscosity flows).

Key Results

  • Second-Order Accuracy: The method achieved second-order convergence in the $L^\infty$ norm for velocity across various Reynolds numbers (up to $Re=1000$).
  • Efficiency: By using quadtrees, the authors achieved results comparable to high-resolution uniform grids but with significantly lower memory and storage requirements.
  • Robustness: The solver successfully handled a "moving flower" domain, maintaining a stable pressure field where other methods might produce "noisy" results due to the changing grid-boundary intersections.
  • 3D Scalability: The method was successfully extended to 3D octree grids, maintaining accuracy in complex volumetric simulations.

Limitations and Future Work

The authors acknowledge a significant hurdle in 3D: the cut-cell approach for the divergence-free condition. In 2D, calculating the "line fraction" of a cell cut by a boundary is straightforward. In 3D, calculating the "surface fraction" of a cell cut by an arbitrary plane involves complex polygonal integration that is difficult to generalize.

Future work will focus on:

  1. Developing a more consistent 3D discretization for the divergence-free equation on irregular domains.
  2. Applying the method to full fluid-structure interaction (FSI) problems, where the fluid and the solid object influence each other's motion.

Summary updated

Abstract

We present a second-order monolithic method for solving incompressible Navier--Stokes equations on irregular domains with quadtree grids. A semi-collocated grid layout is adopted, where velocity variables are located at cell vertices, and pressure variables are located at cell centers. Compact finite difference methods with ghost values are used to discretize the advection and diffusion terms of the velocity. A pressure gradient and divergence operator on the quadtree that use compact stencils are developed. Furthermore, the proposed method is extended to cubical domains with octree grids. Numerical results demonstrate that the method is second-order convergent in $L^\infty$ norms and can handle irregular domains for various Reynolds numbers.
BibTeX
@article{cho2022solving,
  author = {Hyuntae Cho and Yesom Park and Myungjoo Kang},
  title = {{Solving incompressible Navier-Stokes equations on irregular domains and quadtrees by monolithic approach}},
  journal = {Journal of Computational Physics},
  year = {2022},
  url = {https://openreview.net/forum?id=mRdn1c9TqI},
}

Loading related content...

Figure