Abstract
This paper establishes the optimal $H^1$-norm error estimate for a nonstandard finite element method (FEM) designed to approximate $H^2$ strong solutions of second-order linear elliptic PDEs in non-divergence form with continuous coefficients. To overcome the absence of an effective duality argument for this class of non-symmetric operators, a novel analysis technique is introduced. The core of this approach is the establishment of an $H^1$-norm stability estimate for the finite element approximation operator, which mimics the stability properties of the underlying PDE operator. This is achieved through a freezing coefficient technique combined with a topological argument. Notably, the results demonstrate that both $H^1$-norm stability and optimal error estimates hold even for the simplest linear finite element methods.
Introduction
In the field of numerical analysis, the study of elliptic partial differential equations (PDEs) is dominated by the "divergence form" ($\nabla \cdot (A \nabla u) = f$). This form is mathematically convenient because it allows for the use of integration by parts, which simplifies the derivation of weak formulations and the application of standard finite element analysis. However, many critical applications in stochastic optimal control, game theory, and mathematical finance naturally result in PDEs in non-divergence form:
$$-A : D^2u = f$$
When the coefficient matrix $A$ is only continuous rather than smooth, these equations cannot be converted into divergence form. This lack of structure presents a significant challenge for proving the accuracy of numerical approximations. In particular, the standard Aubin-Nitsche duality argument—the primary tool for proving error estimates in lower-order norms like $H^1$—fails because the underlying operators are not symmetric. While previous research established error estimates in the $H^2$-norm for quadratic elements ($r \ge 2$), this paper provides the first rigorous framework to justify the use of simple linear elements ($r=1$) and provides optimal $H^1$ error estimates.
Methodology
The authors analyze a nonstandard $C^0$ finite element method. Their primary innovation is a new analysis technique that avoids the need for a symmetric duality argument by focusing on stability. The approach consists of several key components:
1. Freezing Coefficient Technique
The "freezing coefficient" method relies on the observation that a continuous coefficient matrix $A(x)$ behaves like a constant matrix $A_0$ within a sufficiently small neighborhood. For constant coefficients, the non-divergence operator behaves similarly to a standard divergence operator, which is well-understood. By analyzing the problem locally, the authors can leverage known properties of simpler operators.
2. Discrete $H^1$ Stability
The authors establish a stability estimate for the finite element operator that mirrors the stability of the continuous PDE operator. Specifically, they prove that the $H^1$-norm of the discrete solution is bounded by the $H^{-1}$-norm of the source term:
$$\|u_h\|_{H^1(\Omega)} \lesssim \|f\|_{H^{-1}(\Omega)}$$
This stability is crucial for ensuring that the numerical method does not produce erratic results as the mesh is refined.
3. Topological Argument and Global Estimates
To transition from local "frozen" estimates to a global estimate across the entire domain, the authors employ a covering argument. They use a proof-by-contradiction technique to "strip away" lower-order terms, ensuring the method remains stable globally. This topological approach allows them to handle the non-symmetry of the non-divergence form operator.
4. Linear Elements ($r=1$)
For the specific case of linear elements, where the second derivatives vanish within each element, the authors employ a specialized duality argument. This proves that the method remains invertible and well-posed even when the approximation space lacks the regularity typically associated with $H^2$ solutions.
Key Results
The paper provides several landmark theoretical results for the numerical treatment of non-divergence form PDEs:
-
Optimal $H^1$ Error Estimate: The authors prove that the error between the true solution $u$ and the numerical solution $u_h$ satisfies the optimal bound: $$\|u - u_h\|_{H^1(\Omega)} \lesssim h^l \|u\|_{H^s(\Omega)}$$ where $h$ is the mesh size and $l = \min\{r+1, s\}$. This confirms that the method achieves the "gold standard" of convergence for finite element methods.
-
Stability and Well-Posedness: The research rigorously proves that a unique solution exists for the finite element scheme, even when using linear elements, provided the mesh is sufficiently fine. The stability in the $H^1$-norm ensures the robustness of the approximation.
- Bypassing Symmetry Requirements: By utilizing stability-based analysis rather than symmetry-based duality, the authors provide a new roadmap for analyzing other non-symmetric problems that were previously considered analytically inaccessible.
Conclusion and Future Work
This work fills a significant gap in the numerical analysis of non-divergence form elliptic PDEs. By proving that linear finite elements are stable and provide optimal $H^1$ convergence, the authors justify the use of simpler, more computationally efficient elements for complex problems in finance and control theory.
However, some limitations remain. The current analysis assumes the coefficient matrix $A$ is continuous; extending this to discontinuous ($L^\infty$) coefficients remains an open problem. Additionally, while the proofs hold for $C^{1,1}$ or polygonal domains, more complex geometries require further study. Finally, the authors suggest that this $H^1$ analysis technique could serve as a foundation for analyzing fully nonlinear second-order PDEs, such as the Monge-Ampère or Hamilton-Jacobi-Bellman equations.
Summary updated