Stable finite difference methods for Kirchhoff-Love plates on overlapping grids

Longfei Li, Hangjie Ji, Qi Tang

Abstract

In this work, we propose and develop efficient and accurate numerical methods for solving the Kirchhoff-Love plate model in domains with complex geometries. The algorithms proposed here employ curvilinear finite-difference methods for spatial discretization of the governing PDEs on general composite overlapping grids. The coupling of different components of the composite overlapping grid is through numerical interpolations. However, interpolations introduce perturbation to the finite-difference discretization, which causes numerical instability for time-stepping schemes used to advance the resulted semi-discrete system. To address the instability, we propose to add a fourth-order hyper-dissipation to the spatially discretized system to stabilize its time integration; this additional dissipation term captures the essential upwinding effect of the original upwind scheme. The investigation of strategies for incorporating the upwind dissipation term into several time-stepping schemes (both explicit and implicit) leads to the development of four novel algorithms. For each algorithm, formulas for determining a stable time step and a sufficient dissipation coefficient on curvilinear grids are derived by performing a local Fourier analysis. Carefully designed numerical experiments are carried out to validate the accuracy and stability of the proposed algorithms, followed by two benchmark problems to demonstrate the capability and efficiency of our approach for solving realistic applications.

The Challenge of Complex Geometries

Modeling the vibrations of thin structures, such as aircraft wings or acoustic instruments, typically relies on the Kirchhoff-Love plate theory. Numerically solving this model is challenging due to the presence of the biharmonic operator ($\nabla^4$), a high-order derivative that complicates standard Finite Element Methods (FEM). While Finite Difference (FD) methods offer a simpler alternative, they are traditionally limited to rectangular domains.

To bridge this gap, researchers utilize overlapping grids (or Chimera grids). This approach layers multiple simple, logically rectangular grids to cover complex shapes. However, the numerical interpolation required to transfer information between these overlapping components breaks the symmetry of the underlying physics, introducing "weak instabilities"—small errors that grow exponentially and eventually cause the simulation to fail.

Overlapping Curvilinear Grids

The proposed framework utilizes a collection of boundary-fitted curvilinear grids that overlap with a background Cartesian grid. This allows for the precise representation of complex boundaries, such as holes or irregular edges, without the mesh-generation overhead of unstructured grids.

A composite overlapping grid for a circular hole. The blue Cartesian grid overlaps with a green curvilinear grid, with black dots marking interpolation points.

Stabilization via Hyper-dissipation

The core innovation of this work is the introduction of an artificial hyper-dissipation term. This fourth-order term acts as a high-frequency filter, damping the non-physical noise generated by grid interpolation. Crucially, the authors demonstrate that by carefully scaling the dissipation coefficient, the term remains formally second-order accurate ($O(h^2)$), ensuring that the physical solution for displacement, velocity, and acceleration remains untainted.

Comparison of finite difference stencils for the standard (δ=0) and stabilized (δ=1) schemes, showing the stencil expansion required for the dissipation term.

To implement this, the authors developed four distinct time-stepping schemes:

  • C2 & UPC2: Second-order centered and upwind predictor-corrector schemes.
  • PC22: Based on Adams-Bashforth/Moulton methods.
  • NB2: A predictor-corrector version of the classic Newmark-Beta scheme.

Key Results and Performance

The proposed algorithms were validated through rigorous benchmarks, including manufactured solutions and analytical circular plates. The hyper-dissipation successfully suppressed instabilities across various boundary conditions (free, clamped, and supported).

In practical applications, the efficiency gains were significant. For a plate perforated with 36 holes, the 2D Kirchhoff-Love model required only 1.8 million grid points, compared to the approximately 42 million points that would be necessary for a comparable 3D linear elasticity model. Furthermore, the implicit NB2 scheme allowed for time steps up to ten times larger than explicit methods, making it highly effective for "stiff" structural problems.

Future Directions

While the current framework provides a robust foundation for second-order simulations of isotropic materials, future work will focus on extending these methods to higher-order accuracy and orthotropic materials (such as wood for musical instrument modeling). The ultimate objective is the integration of these stable solvers into complex Fluid-Structure Interaction (FSI) systems.

Summary updated

Abstract

In this work, we propose and develop efficient and accurate numerical methods for solving the Kirchhoff-Love plate model in domains with complex geometries. The algorithms proposed here employ curvilinear finite-difference methods for spatial discretization of the governing PDEs on general composite overlapping grids. The coupling of different components of the composite overlapping grid is through numerical interpolations. However, interpolations introduce perturbation to the finite-difference discretization, which causes numerical instability for time-stepping schemes used to advance the resulted semi-discrete system. To address the instability, we propose to add a fourth-order hyper-dissipation to the spatially discretized system to stabilize its time integration; this additional dissipation term captures the essential upwinding effect of the original upwind scheme. The investigation of strategies for incorporating the upwind dissipation term into several time-stepping schemes (both explicit and implicit) leads to the development of four novel algorithms. For each algorithm, formulas for determining a stable time step and a sufficient dissipation coefficient on curvilinear grids are derived by performing a local Fourier analysis. Quadratic eigenvalue problems for a simplified model plate in 1D domain are considered to reveal the weak instability due to the presence of interpolating equations in the spatial discretization. This model problem is further investigated for the stabilization effects of the proposed algorithms. Carefully designed numerical experiments are carried out to validate the accuracy and stability of the proposed algorithms, followed by two benchmark problems to demonstrate the capability and efficiency of our approach for solving realistic applications. Results that concern the performance of the proposed algorithms are also presented.
BibTeX
@misc{li2021stable,
  author = {Longfei Li and Hangjie Ji and Qi Tang},
  title = {{Stable finite difference methods for Kirchhoff-Love plates on overlapping grids}},
  year = {2021},
  eprint = {2105.05401},
  archivePrefix = {arXiv},
  url = {https://arxiv.org/abs/2105.05401},
}
Figure