Wavelet-based grid adaptation with consistent treatment of high-order sharp immersed geometries

This paper proposes a high-order wavelet-based grid adaptation strategy that utilizes 1D polynomial extrapolation to consistently handle sharp immersed boundaries, thereby enabling robust and predictable error control for solving PDEs on complex, moving geometries.

The Gist

Imagine you are trying to take a high-resolution photograph of a complex scene, like a star-shaped cookie floating in a bowl of soup.

The Problem:
Standard photography (or computer simulations) usually works on a rigid grid, like graph paper. If you want to zoom in on the cookie's jagged edges to see the details, you have to zoom in on the entire bowl of soup, too. This is incredibly wasteful. You end up using millions of pixels just to show the smooth soup, while the interesting action is only happening at the cookie's edge.

Furthermore, if you try to use a special "smart zoom" (called Wavelet-based adaptation) that automatically adds detail only where needed, it usually breaks when it hits the edge of the cookie. The math gets confused because the cookie doesn't line up with the grid lines, causing the "smart zoom" to glitch and lose its accuracy right where you need it most.

The Solution:
The authors of this paper, Changxiao Nigel Shen and Wim M. van Rees, invented a new way to make that "smart zoom" work perfectly, even on jagged, moving, or oddly shaped objects.

Here is how they did it, using some everyday analogies:

1. The "Magic Extrapolation" (Fixing the Glitch)

Normally, when the smart zoom algorithm reaches the edge of the cookie, it tries to guess what the picture looks like just outside the edge. If it guesses wrong (like assuming the soup continues smoothly into the cookie), the math breaks.

The authors' solution is like a skilled artist filling in a missing part of a painting.

• Instead of just guessing blindly, the artist looks at the brushstrokes right next to the edge.
• They also look at the shape of the edge itself (the boundary) and even the direction the edge is curving (derivatives).
• Using this extra information, they "extrapolate" (predict) what the missing pixels should look like so the math stays smooth and consistent.

They call this a High-Order Interpolating Wavelet Transform. In plain English: It's a math trick that lets the computer "pretend" the grid extends smoothly past the edge of the object, so the zoom algorithm doesn't get confused.

2. The "Smart Zoom" (Grid Adaptation)

Once the math is fixed, the computer can finally do what it was meant to do: Adaptive Resolution.

• Where the action is: Near the moving star-shaped cookie, the computer zooms in, using tiny, high-detail grid cells to capture every swirl of the fluid.
• Where it's boring: In the middle of the soup, far from the cookie, the computer zooms out, using large, coarse grid cells.
• The Result: You get a super-detailed simulation of the complex physics without needing a supercomputer. It's like having a camera that automatically focuses only on the subject, leaving the background blurry to save battery.

3. The "Guarantee" (Why it matters)

The most exciting part of this paper is that they didn't just make it work; they proved it works.

• They showed that if you tell the computer, "I want the error to be no bigger than 0.01," the computer will guarantee that the error stays below that limit, no matter how crazy the cookie moves or how complex the soup gets.
• It's like giving a builder a budget and a tolerance level: "Build this house, but make sure the walls are never more than 1 inch off from being straight." The authors proved their method ensures the walls stay within that 1-inch limit, even if the ground shakes.

4. Real-World Applications

They tested this on some tough scenarios:

• A spinning, moving star: Simulating fluid flowing around a star that is both rotating and moving across the screen.
• A vortex hitting a wall: Simulating a swirling tornado hitting a wall at an angle.

In both cases, the method successfully adjusted the grid in real-time, keeping the simulation accurate and efficient.

The Big Picture

Think of this paper as inventing a universal adapter. Before, if you wanted to use high-tech "smart zoom" tools on irregular shapes (like blood vessels, airplane wings, or underwater robots), the tools would break. Now, thanks to this new "extrapolation" trick, those tools work perfectly on any shape, moving or still, ensuring that scientists can simulate complex real-world physics faster and more accurately than ever before.

Technical Summary

1. Problem Statement

The paper addresses the challenge of applying wavelet-based adaptive mesh refinement (AMR) to Partial Differential Equations (PDEs) solved using immersed boundary/interface methods (IBM/IIM).

• The Conflict: Standard wavelet transforms rely on multiresolution analysis to estimate errors and refine grids. They assume the solution field is smooth across the domain. However, immersed methods introduce non-grid-aligned boundaries where field values or derivatives may be discontinuous.
• The Consequence: Standard interpolating wavelet transforms lose their formal order of accuracy near these boundaries. The resulting "detail coefficients" (which act as error indicators) become artificially large due to the discontinuity (Runge phenomenon) rather than physical solution features. This forces the grid to remain uniformly fine near the boundary, negating the efficiency benefits of AMR.
• Current Limitations: Existing approaches often fix the grid resolution near the boundary to a preset value, preventing dynamic adaptation based on solution complexity. Furthermore, no existing wavelet-based strategy preserves the formal wavelet order near sharp, non-grid-aligned interfaces.

2. Methodology

The authors propose a novel high-order interpolating wavelet transform strategy compatible with sharp immersed geometries. The core innovation is a consistent treatment of boundary intersections using polynomial extrapolation.

A. One-Dimensional Framework

• Polynomial Extrapolation: Instead of zero-padding (which causes discontinuities), the method constructs ghost points outside the physical domain using high-order polynomial extrapolation.
• Extrapolation Types:
  • Type I: Uses only interior grid points to extrapolate values outside the domain. Used when the nearest grid point is "even" relative to the boundary.
  • Type II: Incorporates known boundary values (Dirichlet) or derivatives (Neumann) into the polynomial construction. This is used when the nearest grid point is "odd," significantly reducing the Lebesgue constant and mitigating the Runge phenomenon.
• Narrow Intervals (Concave Geometries): In 1D, if the interval between boundaries is too narrow to support the required number of grid points for a standard polynomial, the method switches to a Hermite-like interpolant. This uses both function values and derivatives at the boundary to construct the ghost points.

B. Extension to Multi-Dimensions (2D/3D)

• Tensor Product Approach: The 2D transform is performed via dimension splitting (tensor products of 1D transforms).
• Convex Regions: Standard Type I/II extrapolation is applied along grid lines intersecting the boundary.
• Concave Regions: When a grid line intersects the boundary at two points with insufficient space between them (a "narrow interval"), the method employs multivariate least-squares polynomial fitting.
  • A half-elliptical stencil is defined around control points on the boundary.
  • High-order derivatives are estimated from the interior field using this stencil (similar to Immersed Interface Method discretizations).
  • These estimated derivatives are then used to construct the Hermite-like 1D interpolant required for the narrow interval.

C. Adaptive Grid Strategy

• The method couples the wavelet transform with a temporally adaptive grid strategy.
• Thresholds: A refinement threshold ($\epsilon_r$) and a coarsening threshold ($\epsilon_c$) are defined.
• Error Control: For linear PDEs, the authors prove that the user-defined refinement threshold $\epsilon_r$ provides a rigorous upper bound on the numerical error.
• Algorithm: Every $n$ time steps, the solution is transformed. If detail coefficients exceed $\epsilon_r$, the grid is refined; if they fall below $\epsilon_c$, the grid is coarsened.

3. Key Contributions

1. Consistent High-Order Wavelet Transform: A new algorithm that preserves the formal order $O(h^N)$ of the wavelet transform across arbitrary smooth, irregular domains, including near concave sections and immersed boundaries.
2. Robust Error Bounding: A mathematical proof demonstrating that for linear PDEs, the numerical error is strictly bounded by the user-defined refinement threshold $\epsilon_r$. This establishes a predictable relationship between grid adaptation parameters and solution accuracy.
3. Handling Concave Geometries: A novel approach using multivariate least-squares fitting to estimate boundary derivatives, enabling wavelet transforms in "narrow" concave regions where standard stencils fail.
4. Computational Efficiency: The algorithm maintains a computational complexity of $O(N_p)$ (where $N_p$ is the number of grid points), making it scalable for large simulations.

4. Numerical Results

The method was validated on several test cases using a high-order Immersed Interface Method (IIM) solver:

• Static Field Compression: A sine wave on a star-shaped domain. Results confirmed a linear relationship between the $L_\infty$ error and the compression threshold $\epsilon$, validating the theoretical error scaling.
• Diffusion Equation (Heat Equation): Solved on a star-shaped domain with time-varying Dirichlet conditions. The $L_2$ and $L_\infty$ errors scaled linearly with $\epsilon_r$, confirming the error bound for linear problems.
• Navier-Stokes (Moving Boundary): A translating and rotating star in an incompressible fluid ($Re \approx 774$). The adaptive method captured vortical structures and boundary layers accurately, with errors converging to the reference solution as $\epsilon_r$ decreased.
• Vortex Dipole-Wall Collision: A highly nonlinear benchmark involving a vortex impacting a rotated wall. Despite strong nonlinearities (where the linear error bound proof does not strictly apply), the method successfully suppressed numerical errors below the set threshold and achieved significant grid compression (up to 10x reduction in degrees of freedom) while maintaining accuracy.

5. Significance

• Bridging the Gap: This work successfully bridges the gap between the mathematical rigor of wavelet-based AMR and the geometric flexibility of immersed boundary methods.
• Predictable Accuracy: It provides a "black-box" guarantee for users: setting a refinement threshold $\epsilon_r$ directly controls the maximum numerical error, eliminating the need for trial-and-error grid sizing.
• Efficiency: By allowing dynamic coarsening even near complex boundaries (provided the solution is smooth enough), the method offers substantial computational savings compared to fixed-resolution immersed methods.
• Future Directions: The authors note that extending this to spatially adaptive grids (where resolution varies in space simultaneously) remains a challenge due to interdependencies in the overlap regions, which is identified as a key area for future research.

In summary, this paper presents a robust, mathematically grounded framework for performing high-order wavelet-based grid adaptation on complex, moving geometries, ensuring that numerical errors are rigorously controlled by user-defined parameters.