Discontinuous piecewise polynomial approximation on non-Lipschitz domains

This paper establishes best approximation error estimates for discontinuous piecewise polynomial approximations in fractional Sobolev spaces on non-Lipschitz domains and meshes, even when the boundaries involved are fractal.

The Gist

Imagine you are trying to paint a picture of a very strange, jagged coastline. This isn't a smooth beach; it's a fractal, like the famous "Koch Snowflake." If you zoom in, the edge never gets smooth; it just gets more and more bumpy, with tiny bumps on top of bigger bumps, forever.

Now, imagine you want to use a computer to simulate how sound waves bounce off this coastline. To do this, mathematicians usually break the shape down into small, simple puzzle pieces (like triangles or squares) and approximate the physics on each piece. This is called meshing.

The Problem: The "Smooth" Rules Don't Work

For decades, the rules of the game (mathematical theorems) said: "You can only use puzzle pieces that have smooth, straight edges, and the coastline itself must be smooth enough to fit these pieces."

But what if your coastline is a fractal?

1. You can't fit a smooth triangle perfectly inside a jagged fractal edge.
2. If you try to approximate the fractal with a "pre-fractal" (a slightly smoother version), you introduce huge errors because you're ignoring the tiny, infinite details that actually matter for the physics.

Previously, if you had a fractal domain, you were stuck. You couldn't prove that your computer simulation would get better and better as you added more puzzle pieces.

The Solution: Breaking the Rules (Literally)

This paper by D. P. Hewett introduces a new way of thinking. Instead of forcing the puzzle pieces to be smooth triangles, the author says: "Let the puzzle pieces be as weird as the shape they are trying to fill."

Imagine you are tiling a floor, but the floor has a jagged, fractal edge.

• Old Way: You cut standard square tiles. They don't fit the edge perfectly, leaving gaps or overlapping awkwardly. You have to pretend the edge is smoother than it is.
• New Way (This Paper): You cut tiles that also have fractal edges. One tile might look like a tiny version of the whole snowflake. Another might be a triangle with a fractal side.

The author proves that even if your "tiles" (mesh elements) have fractal boundaries, and even if the "room" (the domain) is a fractal, you can still approximate complex functions with incredible accuracy using discontinuous piecewise polynomials.

What Does "Discontinuous Piecewise Polynomial" Mean?

Think of it like a mosaic made of different colored tiles.

• Piecewise: Each tile has its own simple pattern (a polynomial, like a straight line or a curve).
• Discontinuous: The pattern on one tile doesn't have to match the pattern on the next tile perfectly at the border. They can jump.

In many old methods, the patterns had to flow smoothly from one tile to the next (like a continuous painting). But in this new approach, the author shows that it's okay if the patterns jump at the edges. In fact, allowing this "jump" gives you much more flexibility to handle those crazy, jagged fractal shapes.

The Big Takeaway: "Best Approximation"

The core of the paper is a mathematical guarantee. It says:

"No matter how weird your shape is (even if it's a fractal), and no matter how weird your puzzle pieces are (even if they are fractals), if you make the pieces smaller ($h$) or use more complex patterns on them ($p$), your approximation will get closer to the true answer at a predictable, optimal speed."

Why This Matters

1. Real-World Physics: Many natural phenomena (like sound scattering off a rough rock, or light hitting a complex crystal) happen on shapes that are effectively fractal. This math allows us to simulate them accurately without simplifying the shape too much.
2. No More "Smooth" Assumptions: The author proves that you don't need the shape to be "Lipschitz" (a technical term for "not too jagged"). You can have shapes with infinite jaggedness, and the math still works.
3. Flexibility: It opens the door for engineers and scientists to use "fractal meshes" (meshes made of fractal shapes) to solve problems that were previously too difficult or inaccurate to model.

The Analogy of the "Covering"

To prove this, the author uses a clever trick called a "covering mesh."
Imagine you are trying to measure a jagged coastline. You can't measure the jagged bits directly with a ruler. So, you cover the whole coastline with a grid of large, smooth, overlapping squares.

• You know how to approximate things inside those smooth squares.
• You then prove that because your jagged coastline is inside those smooth squares, and your jagged puzzle pieces fit inside the coastline, the error from the jaggedness is controlled by the error of the smooth squares.

It's like saying: "Even though the coastline is a mess, if I can cover it with a neat grid, and I can solve the problem on the neat grid, I can prove I can solve it on the mess too."

Summary

This paper is a breakthrough because it removes the "smoothness" requirement from the rules of numerical simulation. It tells us that we can model the infinitely complex, jagged world using infinitely complex, jagged puzzle pieces, and we can still trust the math to give us the right answer. It's a permission slip to stop approximating nature as smooth and start modeling it as it really is: rough, fractal, and beautiful.

Technical Summary

Here is a detailed technical summary of the paper "Discontinuous piecewise polynomial approximation on non-Lipschitz domains" by D. P. Hewett.

1. Problem Statement

The paper addresses the challenge of approximating functions using discontinuous piecewise polynomials on non-Lipschitz domains, specifically those with fractal boundaries (e.g., the Koch snowflake).

• Context: Traditional Finite Element Methods (FEM) and Boundary Element Methods (BEM) often rely on conforming meshes with Lipschitz (smooth or polyhedral) elements. When the domain $\Omega$ has a fractal boundary, it cannot be exactly represented by a finite number of Lipschitz elements.
• Current Limitations:
  • Approximating the domain with a "prefractal" (smoother) approximation introduces significant geometric errors.
  • Existing theoretical error estimates (e.g., from [14]) for discontinuous Galerkin (dG) methods assume Lipschitz domains and meshes.
  • There is a lack of rigorous approximation theory for meshes where elements themselves have fractal boundaries, which are necessary to capture the geometry of fractal domains exactly.
• Goal: To establish best approximation error estimates for discontinuous piecewise polynomial spaces on arbitrary meshes of arbitrary domains, without requiring regularity assumptions on the domain boundary or the mesh element boundaries.

2. Methodology

The author employs a combination of local hp-approximation theory, covering mesh techniques, and functional analysis (interpolation and duality).

• Mesh Definition: The paper defines a "mesh" $\mathcal{T}$ as a countable collection of disjoint open subsets of $\Omega$ that cover $\Omega$ up to a set of measure zero. Crucially, these elements can have fractal boundaries.
• Covering Mesh Approach: To handle the lack of regularity, the author introduces a covering mesh $\mathcal{T}^\#$ consisting of standard geometric shapes (simplices or parallelotopes) that contain the mesh elements.
  • A choice function $\kappa: \mathcal{T} \to \mathcal{T}^\#$ maps each mesh element $K$ to a covering element $\tilde{K}$.
  • This allows the application of standard polynomial approximation results on the regular covering elements $\tilde{K}$ to bound errors on the irregular mesh elements $K$.
• Function Spaces: The analysis is conducted in both:
  • Intrinsic spaces: $W^m(\Omega)$ (Sobolev spaces defined via weak derivatives on $\Omega$).
  • Extrinsic spaces: $H^s(\Omega)$ (Restrictions of Bessel potential spaces $H^s(\mathbb{R}^n)$) and $\tilde{H}^s(\Omega)$ (Closure of $C_0^\infty(\Omega)$ in $H^s(\mathbb{R}^n)$).
• Key Technical Tools:
  • Local hp-estimates: Utilizing optimal approximation operators on the covering elements.
  • Interpolation Theory: Using the K-method to derive estimates for fractional order spaces ($0 \le s \le m$).
  • Duality Arguments: Extending results to negative order norms and distributions.

3. Key Contributions

The paper provides a comprehensive theoretical framework that generalizes previous results to the most general setting possible:

1. Generalization to Non-Lipschitz Settings: The results hold for arbitrary domains and arbitrary meshes, including those with fractal boundaries or elements with positive Lebesgue measure boundaries. No regularity assumptions (like Lipschitz continuity) are required on the domain or mesh geometry.
2. Strengthening of Local Estimates (Theorem 2.3):
   • Generalizes [14, Lemma 4.31] to the non-Lipschitz case.
   • Crucial Improvement: Unlike previous works, this theorem aggregates the error contributions of all mesh elements associated with a single covering element into a single estimate.
   • Consequence: This aggregation eliminates the need for a "mesh density" assumption (limiting how many mesh elements can intersect a covering element) when deriving global error estimates.
3. Global Error Estimates (Theorem 2.6):
   • Provides global best approximation error estimates in broken Sobolev norms for functions in extrinsic spaces $H^m(\Omega)$.
   • The bounds depend on the mesh width $h$, polynomial degree $p$, and the regularity of the function, with constants independent of the domain's geometry.
4. Extension to Intrinsic and Fractional Spaces:
   • Corollary 2.7: Adapts results to intrinsic spaces $W^m(\Omega)$ under the assumption that $\Omega$ is a $W^m$-extension domain (e.g., Koch snowflake).
   • Corollary 2.8: Derives $L^2$ estimates for fractional order spaces $H^s(\Omega)$ ($0 \le s \le m$).
   • Corollary 2.9 & 2.10: Extends estimates to negative order norms and distributions, provided the space family forms an interpolation scale.

4. Main Results (Theorems and Corollaries)

• Theorem 2.3 (Local hp-estimate): For a function $u$ locally in $W^{m_K}$, the error of the best polynomial approximation of degree $p$ on a mesh element $K$ is bounded by $O(h_K^{t-j} (p+1)^{-(m_K-j)})$, where $t = \min(m_K, p+1)$. The constant depends only on dimension, shape regularity of the covering element, and $h_0$ (max covering size), not on the specific geometry of the fractal boundary.
• Theorem 2.6 (Global Extrinsic Estimate): For $u \in H^m(\Omega)$, the global broken Sobolev norm error is bounded similarly, with constants depending only on $n, m, h_0$.
• Corollary 2.8 (Fractional Order): For $u \in H^s(\Omega)$ with $0 \le s \le m$, the$L^2$error is$O((h/(p+1))^s)$.
• Corollary 2.9 (Negative Order): Provides estimates for the error in $\tilde{H}^{s_1}(\Omega)$ norms for $s_1 \le 0$, bounded by the $H^{s_2}(\Omega)$ norm of the function ($s_2 \ge 0$).
• Corollary 2.10 (Distributions): Extends negative order estimates to distributions, requiring the interpolation scale property (satisfied by $(\epsilon, \delta)$ locally uniform domains like the Koch snowflake).

5. Significance and Applications

• Rigorous Foundation for Fractal Numerics: This work provides the first rigorous theoretical justification for using discontinuous piecewise polynomial approximations on meshes that exactly capture fractal geometries.
• Integral Equation Methods: The results are directly motivated by and applicable to Boundary Element Methods (BEM) and Volume Integral Equation methods for scattering problems on fractal screens and inhomogeneities (e.g., Lippmann-Schwinger equation). Previous work used piecewise constant approximations; this paper extends the theory to higher-order polynomials.
• Discontinuous Galerkin (FEM): The results support the development of dG-FEM on fractal domains, allowing for higher-order convergence rates without the geometric approximation errors inherent in prefractal meshing.
• Relaxation of Geometric Constraints: By removing the Lipschitz requirement, the paper opens the door to analyzing numerical methods on a much broader class of physical and mathematical domains, including those with cusps, slits, and self-similar fractal structures.

In summary, Hewett successfully bridges the gap between classical approximation theory and the complex geometry of fractal domains, proving that optimal convergence rates can be achieved even when the domain and mesh elements possess fractal boundaries, provided the approximation space is discontinuous.