Error estimates for the patch bubble method for convection-dominated channel flow problem

1. Problem Statement

The paper addresses the numerical solution of the convection-dominated advection-diffusion equation in a two-dimensional square domain $\Omega = (0,1)^2$. The problem is defined as:
$$-\epsilon \Delta u_\epsilon + \mathbf{a} \cdot \nabla u_\epsilon = f \quad \text{in } \Omega, \quad u_\epsilon = 0 \quad \text{on } \partial\Omega$$
where:

• $0 < \epsilon \ll 1$ is the diffusion coefficient (convection-dominated regime).
• $\mathbf{a} = (-1, 0)$ is the constant advection velocity (parallel flow).
• $f \equiv 1$ is the source term.

Challenges:
In the convection-dominated regime ($\epsilon \ll h$, where $h$ is the mesh size), standard Galerkin finite element methods suffer from spurious oscillations and lack of stability near boundary layers. The solution exhibits:

1. An elliptic boundary layer of width $\mathcal{O}(\epsilon)$ at the outflow boundary ($x=0$).
2. Two parabolic boundary layers of width $\mathcal{O}(\sqrt{\epsilon})$ at the top and bottom boundaries ($y=0, y=1$).
3. Corner layers where these boundaries intersect.

2. Methodology: The BMZ Method

The authors analyze the Bubble Mesh Zoom (BMZ) method, a residual-free bubble (RFB) approach that enriches the standard finite element space.

2.1 Discretization Space

The discrete space $V_h$ is constructed as $V_h = V_L \oplus V_B$, where:

• $V_L$ (Coarse Space): Standard bilinear finite elements ($Q_1$) vanishing on the boundary.
• $V_B$ (Bubble Space): Enriched with two types of "residual-free bubbles" that solve the local differential equation exactly on sub-domains:
  1. Element Bubbles ($B_T$): Defined on each element $T$.
  2. Patch Bubbles ($B_S$): Defined on "patches" $\omega_S$ consisting of two adjacent elements sharing an edge $S$.

2.2 Key Innovation: Patch Bubbles

Unlike classical RFB methods that only use element bubbles, BMZ introduces bubbles on patches.

• Rationale: Element bubbles alone often fail to capture the interaction between boundary layers and the flow direction accurately, leading to oscillations and "corner spikes."
• Implementation: The bubbles are computed by solving the local advection-diffusion problem on the patch with a specific right-hand side (residual) derived from the coarse solution.
• Computation: The authors employ a recursive algorithm ("Zooming") to compute these bubbles. If the local mesh size is larger than the diffusion scale ($h > \epsilon$), the problem is solved on a finer sub-mesh using the same bubble method. This recursion continues until $h < \epsilon$, where a standard Galerkin method is used.

3. Theoretical Analysis and Main Contributions

3.1 Asymptotic Expansion

The proof relies on an asymptotic expansion of the exact solution $u_\epsilon$ (based on work by Gie, Jung, and Temam):
$$u_\epsilon \approx u_{as} = u_0 + \text{correctors}$$

• $u_0 = 1-x$ is the zeroth-order convective solution.
• Correctors: Functions ($\phi, \xi, \theta, \zeta$) are introduced to compensate for boundary layer behaviors (parabolic, elliptic, and corner layers).

3.2 Construction of the Interpolant

The authors construct a specific interpolant $u_I = u_L + u_B$ within the discrete space $V_h$ to approximate $u_{as}$.

• $u_L$: Standard bilinear interpolation of $u_0$ in the interior.
• $u_B$: A linear combination of bubble functions with carefully chosen coefficients ($\alpha_\lambda$).
  • In parabolic layers ($y=0, 1$), coefficients are chosen to cancel the error $u_0 - u_L$ exactly using a superposition of element and patch bubbles.
  • In the elliptic layer ($x=0$), coefficients are chosen to ensure the bubble's "theta-corrector" cancels the exact solution's theta-corrector, while minimizing other error terms.

3.3 Main Theoretical Result (Theorem 6)

The paper proves an error estimate in a specific energy norm $|||\cdot|||$:
$$|||u_\epsilon - u_h||| \leq C \left(\frac{\epsilon}{h}\right)^{1/4}$$
Conditions:

• Valid for the convection-dominated regime where $0 < \epsilon/h \leq \hat{\epsilon}_0$ (with $\hat{\epsilon}_0$ small).
• The constant $C$ is independent of $\epsilon$ and $h$.

Significance of the Estimate:

• Previous estimates for similar methods (e.g., RFBe by Cangiani and Süli) often degenerate or blow up when the flow aligns with coordinate axes.
• The BMZ estimate remains robust for parallel flow ($\mathbf{a} = (-1,0)$).
• The rate $(\epsilon/h)^{1/4}$ is derived for a norm stronger than those used in previous literature, providing a more rigorous guarantee of stability.

4. Numerical Results

The authors validate the theory with numerical experiments comparing BMZ, Classical RFB, and Enhanced RFB (RFBe).

• Qualitative Behavior:

  • RFB: Exhibits visible oscillations near the boundary layer and a "corner spike" (max value $\approx 1.41$ vs. expected $\approx 1$).
  • RFBe: Smoother but still requires multiple element layers to resolve the boundary layer decay.
  • BMZ: Eliminates oscillations and corner spikes entirely. The maximum value is $\approx 0.984$ (very close to 1). It resolves the $\epsilon$-width layer within a single element layer.

• Quantitative Convergence (Tables 1-3):

  • $H^1$-seminorm: RFB and RFBe show nearly constant errors (no convergence) as the mesh refines. BMZ shows clear convergence with an Experimental Order of Convergence (EOC) $\approx 1.2 - 1.3$.
  • Stability Norm: BMZ achieves the smallest error for fine meshes ($N=160$), roughly half the error of RFBe and an order of magnitude better than RFB. EOC for BMZ ranges from 1.1 to 2.0.
  • $L^2$ Norm: BMZ converges significantly faster than the other methods, eventually becoming the most accurate despite starting with a higher error on coarse meshes.

5. Conclusion and Significance

• Robustness: The BMZ method successfully handles convection-dominated flows with parallel wind directions, a case where previous patch-bubble methods (like RFBe) theoretically fail or perform poorly.
• Accuracy: By utilizing patch bubbles, the method captures the essential physics of boundary layers (both parabolic and elliptic) more effectively than element-only bubbles.
• Theoretical Breakthrough: The derivation of the $(\epsilon/h)^{1/4}$ error estimate provides the first rigorous theoretical justification for the superior performance of patch-based bubble methods in parallel flow scenarios.
• Practicality: The recursive "zoom" algorithm allows for the practical computation of these complex residual-free bubbles without requiring prohibitively fine global meshes.

In summary, this paper establishes that enriching finite element spaces with patch-based residual-free bubbles provides a theoretically sound and numerically superior stabilization technique for convection-diffusion problems, effectively eliminating spurious oscillations and achieving optimal convergence rates in the convection-dominated regime.