Polynomial quasi-Trefftz DG for PDEs with smooth coefficients: elliptic problems

This paper introduces a polynomial quasi-Trefftz discontinuous Galerkin method for variable-coefficient elliptic problems that utilizes Taylor polynomial-based approximate solutions to achieve higher accuracy and high-order convergence compared to standard DG schemes, while also addressing non-homogeneous sources through local particular solutions.

The Gist

Imagine you are trying to paint a massive, complex mural on a wall that has a strange, shifting texture. The wall represents the physical world (like heat flowing through a metal plate or wind blowing through a city), and the "texture" is the math describing how things move and change.

To paint this mural, you need a team of artists (mathematicians) and a set of brushes (algorithms).

The Old Way: The "Blank Canvas" Approach

Traditionally, mathematicians use a method called Discontinuous Galerkin (DG). Think of this as using a standard set of brushes that can paint any shape: straight lines, curves, waves, whatever. These are called "polynomials."

To get a very detailed, high-quality picture, you need a huge number of these standard brushes. You have to use thousands of tiny strokes to approximate the complex curves of the wall. It works, but it's slow and computationally expensive because you are carrying around a massive toolbox of generic tools, most of which aren't perfectly suited for the specific job at hand.

The New Idea: The "Custom-Molded" Approach

This paper introduces a smarter way: Polynomial Quasi-Trefftz DG.

Instead of using generic brushes, imagine you could create a custom brush for every single section of the wall. This custom brush is shaped exactly like the curve the wall wants to be in that specific spot.

• Trefftz Methods: If the wall's texture was simple and unchanging (like a flat piece of wood), you could easily make a brush that fits perfectly. This is the "Trefftz" method. It's incredibly efficient because you need very few strokes to get a perfect picture.
• The Problem: Real-world walls (PDEs with variable coefficients) are messy. The texture changes from spot to spot. You can't make a single perfect brush because the "exact solution" (the perfect curve) is too hard to find for every spot.
• The Solution (Quasi-Trefftz): This paper says, "We can't find the perfect brush, but we can make a 'Quasi' brush."
  • A Quasi-Trefftz brush is a "good enough" approximation. It's a polynomial that mimics the wall's behavior so closely that the error is tiny—so tiny it's almost invisible.
  • It's like using a cookie cutter that isn't exactly the shape of the dough, but is so close that when you press it down, the result looks perfect to the naked eye.

How It Works (The Magic Recipe)

The authors developed a recipe (an algorithm) to bake these custom brushes:

1. Look at the Wall: They analyze the specific rules of the wall (the equation) at a specific point.
2. The "Cauchy Data" (The Seed): They pick a starting point and decide on a few initial "seed" values (like the slope and height at that one spot).
3. The Algorithm: Using a clever step-by-step process, they calculate the rest of the brush shape. They don't need to solve the whole wall at once; they just build the brush locally, piece by piece.
4. The Result: They end up with a set of custom brushes that are much smaller and more efficient than the generic ones.

Why Is This a Big Deal?

The paper proves two amazing things:

1. Same Quality, Less Effort: You get the same high-quality picture (accuracy) as the old method, but you need far fewer brushes (degrees of freedom). It's like painting a masterpiece with 100 custom strokes instead of 10,000 generic ones.
2. Speed: Because you have fewer things to calculate, the computer finishes the job much faster. The paper shows that for complex 3D problems, the new method is significantly quicker.

The "Smooth" Requirement

There is one catch: The wall needs to be relatively smooth (the coefficients of the equation must be smooth). If the wall is jagged and broken (like a corner with a sharp singularity), the custom brushes might struggle a bit. However, the authors show that even in tricky situations (like wind blowing around a sharp corner), the method still performs very well.

The Bottom Line

This paper is like inventing a 3D printer for mathematical solutions. Instead of chiseling away at a block of stone with a generic hammer (standard methods), you print a custom tool that fits the shape of the problem perfectly.

• For Scientists: It means solving complex physics problems (like heat diffusion or fluid flow) faster and with less computer power.
• For Everyone: It's a reminder that sometimes, the best way to solve a hard problem isn't to throw more brute force at it, but to understand the problem deeply enough to build a tool that fits it perfectly.

The authors have even made their "3D printer" (the code) available for free, so other scientists can start using these custom brushes immediately.

Technical Summary

Here is a detailed technical summary of the paper "Polynomial quasi-Trefftz DG for PDEs with smooth coefficients: elliptic problems."

1. Problem Statement

The paper addresses the numerical solution of linear partial differential equations (PDEs) with smooth, variable coefficients, specifically focusing on elliptic diffusion–advection–reaction problems.

• Context: Classical Discontinuous Galerkin (DG) methods use piecewise polynomial spaces that approximate all regular functions but are not tailored to the specific PDE, leading to a high number of degrees of freedom (DOFs) for high accuracy.
• Limitation of Standard Trefftz Methods: Standard Trefftz methods use basis functions that are exact solutions to the PDE on each element. While highly efficient for PDEs with piecewise-constant coefficients (e.g., Laplace, Helmholtz), they fail for variable coefficients because exact local solutions are generally unavailable.
• Goal: To develop a method that retains the high accuracy and reduced DOF count of Trefftz methods for PDEs with smooth variable coefficients and non-zero source terms.

2. Methodology

A. Polynomial Quasi-Trefftz Spaces

The authors define a Polynomial Quasi-Trefftz space, denoted $QT^p_f(E)$, for a general linear PDE operator $M$ of order $m$ with smooth coefficients $\alpha_j$ and source term $f$.

• Definition: A polynomial $v \in P_p(E)$ belongs to $QT^p_f(E)$ if it satisfies the PDE "up to a small residual" at a specific point $x_E$ within the element $E$. Specifically, the derivatives of the residual $Mv - f$ vanish at $x_E$ up to order $p-m$:
  $$D^i(Mv - f)(x_E) = 0 \quad \forall |i| \leq p-m$$
• Approximation Properties: The paper proves (Theorem 2.4) that these spaces approximate smooth solutions with the same convergence rates ($O(h^{p+1})$ in $L^2$ and $O(h^p)$ in energy norms) as the full polynomial space $P_p(E)$, despite having significantly fewer dimensions.
• Dimension Reduction: For a second-order PDE ($m=2$) in $d$ dimensions, the dimension of the quasi-Trefftz space scales as $O(p^{d-1})$, whereas the full polynomial space scales as $O(p^d)$. This results in a substantial reduction in DOFs for high polynomial degrees.

B. Construction Algorithm

The authors provide Algorithm 1 to construct a basis for these spaces:

• Non-degeneracy Condition: The method requires a mild condition that the coefficient of the highest-order derivative in a specific direction (e.g., $\alpha_{me_1}$) is non-zero at $x_E$. This is naturally satisfied for elliptic operators.
• Cauchy Data: The construction relies on specifying "Cauchy data" (values of the first $m$ derivatives) on a hyperplane within the element.
• Iterative Process: The algorithm iteratively computes the coefficients of the monomial expansion of the basis functions using the PDE coefficients and their derivatives at $x_E$. This process is completely parallelizable across elements.
• Non-homogeneous Problems: For $f \neq 0$, the space is affine. The method constructs a particular approximate solution $u_{h,f}$ (using the same algorithm with arbitrary Cauchy data, typically zero) and solves for the difference $u_{h,0}$ in the homogeneous quasi-Trefftz space.

C. Discontinuous Galerkin Formulation

The authors formulate a DG method for the diffusion–advection–reaction equation:

• Diffusion: Handled via the Symmetric Interior Penalty (SIPG) method.
• Advection-Reaction: Handled via an upwind flux formulation.
• Stability Analysis: The paper proves well-posedness, stability, and quasi-optimality for general discrete subspaces, adapting standard DG analysis to handle the specific structure of quasi-Trefftz spaces (which do not necessarily contain their own derivatives).
• Convergence: Under the assumption that the exact solution is sufficiently smooth ($C^{p+1}$), the method achieves optimal $h$-convergence rates.

3. Key Contributions

1. Generalization of Quasi-Trefftz Methods: Extends the quasi-Trefftz concept from oscillatory problems (wave/Helmholtz) to general elliptic diffusion–advection–reaction problems with smooth variable coefficients.
2. Rigorous Approximation Theory: Proves that polynomial quasi-Trefftz spaces retain the optimal approximation power of full polynomial spaces while drastically reducing the dimension.
3. Constructive Algorithm: Provides a simple, parallelizable iterative algorithm (Algorithm 1) to compute bases for these spaces given smooth coefficients.
4. Stability and Convergence Proof: Establishes stability and optimal error bounds for the DG scheme applied to these spaces, handling the non-nested nature of the spaces and the affine trial space for non-homogeneous problems.
5. Implementation: The method is implemented in NGSolve (using the NGSTrefftz library), demonstrating practical feasibility.

4. Numerical Results

The paper presents numerical experiments in 2D and 3D:

• Accuracy: The quasi-Trefftz DG method achieves error levels and convergence rates ($O(h^{p+1})$ in $L^2$, $O(h^p)$ in energy norm) comparable to standard DG methods using full polynomial spaces.
• Efficiency:
  • DOF Reduction: For the same accuracy, the quasi-Trefftz method uses significantly fewer DOFs (e.g., for $p=4$ in 3D, the reduction is substantial).
  • Computational Time: Despite the overhead of computing basis functions, the total time (assembly + solve) is lower for the quasi-Trefftz method when the mesh is fine or $p$ is high, due to the smaller linear systems.
• Advection-Dominated Regimes: The method performs robustly in advection-dominated problems (low diffusion $\nu$), capturing internal layers and boundary layers without significant spurious oscillations, comparable to standard DG.
• Conditioning: While the condition numbers grow with $p$, they remain manageable. The authors note that optimizing the choice of Cauchy data could further improve conditioning.

5. Significance

This work bridges a critical gap in high-order numerical methods:

• It enables the use of Trefftz-like efficiency (high accuracy with few DOFs) for a much broader class of PDEs (variable coefficients) where exact Trefftz bases are impossible.
• It offers a computationally superior alternative to standard DG for smooth problems, reducing memory usage and solver time without sacrificing accuracy.
• It provides a systematic framework for constructing these spaces, moving beyond ad-hoc approaches for specific equations.
• The method is particularly promising for problems where high-order accuracy is required, as the dimension reduction becomes more pronounced with increasing polynomial degree $p$.

Future Directions: The authors suggest optimizing the choice of Cauchy data for better conditioning, extending the analysis to Sobolev norms (weaker regularity), investigating $p$-convergence rates, and applying the method to mixed-type PDEs (e.g., Euler-Tricomi).