Characteristic boundary conditions for Hybridizable Discontinuous Galerkin methods

This paper introduces characteristic boundary conditions, including Navier-Stokes characteristic boundary conditions and a novel generalized characteristic relaxation approach, within the Hybridizable Discontinuous Galerkin framework to effectively minimize wave and vortex reflections at artificial boundaries in both inviscid and viscous weakly compressible flows.

The Gist

Imagine you are trying to record a symphony in a small, echoey room. You want to capture the music perfectly, but the walls are too close. Every time a sound wave hits a wall, it bounces back (an echo), ruining your recording. In the world of computer simulations for air and fluid flow, this is a massive problem. Scientists want to simulate how air moves around a plane or a car, but they can't simulate an infinite sky. They have to cut off the simulation at a certain point (an "artificial boundary").

The problem? When sound waves or swirls of air (vortices) hit this artificial wall, they usually bounce back into the simulation, creating fake noise and errors. It's like trying to film a movie in a room where the walls keep throwing the actors back onto the set.

This paper introduces a clever new way to build these "walls" so they act like ghosts instead of mirrors. They let the air and sound pass through as if the wall wasn't there at all.

Here is the breakdown of their solution using simple analogies:

1. The Problem: The "Hard Wall" vs. The "Ghost Wall"

In standard computer simulations, the boundary is often a "hard wall." If a wave hits it, it bounces back.

• The Old Way (Hard Wall): Imagine shouting at a concrete wall. The sound bounces back at you. In a simulation, this "echo" messes up the data, making it look like the air is reacting to things that aren't real.
• The Goal (Ghost Wall): We want a wall that acts like a window. When a wave hits it, it should just walk right through and disappear, never to be seen again.

2. The New Tool: "Characteristic Boundary Conditions" (CBCs)

The authors developed a special set of rules for these walls, called Characteristic Boundary Conditions. Think of this as giving the wall a "smart brain" that understands the language of waves.

Instead of just saying "Stop!" or "Reflect!", the wall listens to the waves and asks: "Are you coming in or going out?"

• If a wave is leaving: The wall says, "Go ahead, I won't stop you."
• If a wave is trying to enter: The wall says, "Wait, let's check what the outside world is doing and adjust you so you don't cause a crash."

3. The Two Types of "Smart Walls"

The paper introduces two specific versions of this smart wall:

A. The "Classic Smart Wall" (NSCBC)

This is an upgrade to an existing method. It's like a traffic cop who knows the speed limit.

• It knows that sound waves travel at specific speeds.
• It gently nudges the air pressure at the edge to match the "target" pressure outside, so the wave doesn't feel a sudden stop and bounce back.
• The Catch: It requires the user to guess a few numbers (like how hard to nudge the traffic). If you guess wrong, the wall might still reflect a little bit.

B. The "Super Smart Wall" (GRCBC) - The Novelty

This is the big new invention in the paper. It's like a self-driving car that adjusts its own steering.

• The Problem with the Old Way: In complex computer grids (like a mesh of tiny triangles), it's hard to know exactly how far away the "next" point is to calculate the nudge.
• The Solution: The authors created a method that uses the computer's own time-step (how fast the simulation runs) to figure out the perfect nudge automatically.
• The Benefit: You don't need to guess the numbers anymore. The wall figures out the perfect amount of "relaxation" (how soft the wall should be) based on the math of the simulation itself. It's like having a wall that automatically adjusts its transparency based on the weather.

4. Why This Matters: The "Vortex" Test

The authors tested their new walls with some tricky scenarios:

• The Sound Pulse: A simple sound wave traveling through the air. The new walls let it pass perfectly without an echo.
• The Swirling Vortex: Imagine a tornado or a swirl of air moving toward the edge of the simulation. This is the hardest test.
  • Old walls: The swirl hits the wall, bounces back, and creates a mess of fake turbulence.
  • New walls: The swirl hits the wall, and the wall "absorbs" it, letting it exit the simulation cleanly.

5. The "HDG" Connection

The paper uses a specific type of math called Hybridizable Discontinuous Galerkin (HDG).

• Analogy: Imagine building a house out of Lego bricks.
  • Old methods: You have to glue every single brick together tightly. If you want to change one brick, you have to redo the whole house. It's slow and heavy.
  • HDG: You use special connectors that let the bricks talk to each other without being glued. You can swap out a whole section of the wall easily.
• The authors managed to fit their "Smart Wall" rules right into these Lego connectors. This means the simulation runs faster and uses less computer power while still being incredibly accurate.

The Big Takeaway

This paper is like inventing a soundproof window for computer simulations of air and water.

• Before, if you wanted to simulate a plane flying, you had to make your simulation box huge to push the "echoes" far away, which took forever to compute.
• Now, with these new "Smart Walls," you can have a smaller simulation box, and the air will just flow out the back without bouncing back in.

In short: They taught the computer's "walls" how to be polite ghosts, letting the air and sound pass through without throwing a tantrum and ruining the show. This makes simulations of flight, wind, and sound much faster, cheaper, and more accurate.

Technical Summary

1. Problem Statement

The direct numerical simulation (DNS) of unsteady compressible flows, particularly in computational aeroacoustics (DNC), requires high-order, low-dissipative numerical methods to resolve both aerodynamic and acoustic scales. A critical challenge in these simulations is the truncation of the computational domain, which necessitates the use of artificial boundary conditions (ABCs) to mimic an infinite domain.

Standard boundary conditions often cause spurious wave reflections at artificial boundaries, contaminating the solution. While Characteristic Boundary Conditions (CBCs)—based on the characteristic decomposition of the Euler/Navier-Stokes equations—are widely used in Finite Difference (FD) methods to minimize these reflections, their implementation in Finite Element (FE) methods, specifically Discontinuous Galerkin (DG) and Hybridizable Discontinuous Galerkin (HDG), is non-trivial.

• The Gap: In DG/HDG, boundary conditions are enforced weakly via numerical fluxes requiring an explicit external state ($U^-$). Previous DG implementations of CBCs relied on ad-hoc solutions like "mirror elements" or least-squares approaches using nodal shape functions, which complicate the implementation and increase computational overhead.
• The Goal: To integrate robust CBCs (specifically Navier-Stokes CBCs and Generalized Characteristic Relaxation CBCs) directly into the HDG framework without auxiliary structures, ensuring minimal reflection of acoustic waves and vortices for weakly compressible flows ($Ma_\infty \leq 0.3$).

2. Methodology

2.1. Theoretical Framework

The authors start from the compressible Navier-Stokes equations and derive the characteristic quasi-linear form of the Euler equations.

• Characteristic Decomposition: The system is decomposed into incoming and outgoing waves based on eigenvalues of the convective Jacobian ($\Lambda = \text{diag}(u_n-c, u_n, u_n, u_n+c)$).
• NSCBCs (Navier-Stokes Characteristic BCs): These model incoming wave amplitudes ($L^-$) by relaxing them toward a target state ($U^-$) using a relaxation matrix ($D^-$) and a characteristic length ($L$). They also incorporate transverse convective terms ($T^-$) and viscous terms ($V$) to handle multidimensional effects.
• GRCBCs (Generalized Characteristic Relaxation BCs): A novel approach introduced in this work. Instead of relying on a user-defined characteristic length $L$ (which is ambiguous in FE meshes), GRCBCs define the relaxation matrix based on the CFL condition. The relaxation matrix is derived as $D^- \sim \frac{1}{\Delta t} C$, where $C$ is a user-tunable diagonal matrix of relaxation factors ($0 < C_i \leq 1$). This allows the method to adapt to the time step and mesh resolution naturally.

2.2. HDG Implementation

The authors propose a unified formulation for the HDG method that avoids mirror elements:

1. Weak Enforcement: Boundary conditions are imposed via the numerical flux boundary operator $\hat{\Gamma}_h$.
2. Filtering Mechanism: The authors derive a boundary operator that filters outgoing waves (determined by the interior state) and models incoming waves (determined by the target state and relaxation).
3. Stabilization: A modified far-field operator is proposed to handle cases where eigenvalues vanish (e.g., $u_n=0$), ensuring numerical stability by premultiplying with the inverse of the absolute Jacobian.
4. Viscous and Transverse Terms: The formulation explicitly includes transverse derivatives ($\partial/\partial \eta$) and viscous fluxes ($V$), which are crucial for accurately capturing vortex interactions and non-planar wave propagation.

The resulting boundary operator $\hat{\Gamma}_h$ takes the form:
$$\hat{\Gamma}_h = \hat{A}^+(\hat{U}_h - U_h) - \hat{A}^-\left( \hat{U}_h - \hat{U}_h^n + \Delta t \left[ \hat{P} \hat{D} \hat{P}^{-1}(\hat{U}_h - U^-) + \beta \hat{B} \frac{\partial \hat{U}}{\partial \eta} - \hat{V} \right] \right)$$
where $\hat{A}^\pm$ are incoming/outgoing Jacobians, $\beta$ is a transverse relaxation parameter, and $U^-$ is the target state.

3. Key Contributions

1. First HDG Implementation of CBCs: The paper presents the first derivation and implementation of both NSCBCs and the novel GRCBCs within the HDG framework.
2. Novel GRCBC Formulation: The introduction of GRCBCs eliminates the need for user-defined characteristic lengths or mirror elements. By linking the relaxation matrix to the CFL condition and time step, the method becomes more robust and easier to tune.
3. Unified Framework: The proposed method recovers common HDG boundary conditions (Subsonic Outflow, Far-Field) as special cases, providing a flexible spectrum from "hard" boundaries to "perfectly non-reflecting" boundaries via the tuning of relaxation factors.
4. Viscous and Transverse Handling: The implementation rigorously incorporates transverse convective terms and viscous boundary conditions (zero normal gradients for stress/heat flux), which are often neglected in simpler 1D-based CBC implementations.

4. Numerical Results

The authors validated the method using four benchmarks in 2D, comparing results against reference solutions (computed on larger domains with sponge layers) and standard HDG boundary conditions (Subsonic Outflow - SO, Far-Field - FF).

• Planar Acoustic Pulse (1D):
  • Finding: The Far-Field target state ($U_\infty$) acts as a perfectly non-reflecting boundary for planar waves regardless of relaxation tuning.
  • Finding: Hard target states (e.g., fixed pressure $p_\infty$) cause reflections unless the relaxation matrix is tuned to zero (perfectly non-reflecting).
• Oblique Pressure Pulse (2D Non-linear):
  • Finding: For non-linear pulses, the NSCBC with the subsonic outflow target state ($U_{p_\infty}$) and optimal relaxation parameter ($\sigma \approx 0.28$) yielded the lowest reflection.
  • Finding: Far-field target states ($U_\infty$) showed slight overdamping due to the lack of transverse terms but remained robust against flow inversion.
• Zero-Circulation Vortex (Vortex Convection):
  • Finding: This is the most critical test. Standard SO and FF conditions caused significant reflections when vortices crossed the boundary.
  • Result: The proposed CBCs (specifically GRCBC/NSCBC with $U_{p_\infty}$ target state + transverse terms $\beta$) significantly minimized vortex reflections. The inclusion of transverse terms was essential for accuracy.
• Laminar Flow Around a Cylinder (Viscous, Vortex Shedding):
  • Finding: In a viscous, unsteady flow generating dipole sound, standard boundary conditions generated upstream-propagating disturbances.
  • Result: The GRCBC/NSCBC with $U_{p_\infty}$ and viscous terms reduced pressure disturbances across the domain significantly compared to SO and FF, while maintaining accurate aerodynamic coefficients (drag/lift).

5. Significance and Conclusion

The paper demonstrates that Characteristic Boundary Conditions can be successfully and efficiently integrated into HDG methods without the computational overhead of mirror elements or complex shape-function manipulations.

• Superiority: The proposed CBCs, particularly the GRCBC variant, outperform traditional HDG boundary conditions (SO and FF) in minimizing spurious reflections of both acoustic waves and vortices at artificial boundaries.
• Flexibility: The GRCBC approach offers a tunable parameter (relaxation factor $C$) that allows users to balance between strict target enforcement and non-reflecting behavior.
• Practical Impact: For aeroacoustic simulations involving weakly compressible flows, the authors conclude that the new CBCs should be preferred over standard methods, especially at subsonic outflow boundaries where vortex shedding occurs.

The work bridges a significant gap between the theoretical robustness of characteristic methods and the practical implementation requirements of high-order finite element methods.