Abstract
For over 25 years, Essentially Non-Oscillatory (ENO) methods have been a cornerstone of computational fluid dynamics, yet their global stability lacked a rigorous mathematical proof. This paper addresses this gap by proving that ENO reconstruction and interpolation are stable across any order of accuracy and non-uniform meshes. By establishing a "sign property"—where reconstructed jumps match the direction of underlying data—and deriving bounded stability constants, the authors provide a formal foundation for the long-standing empirical success of ENO schemes.
Introduction
Simulating phenomena with sharp discontinuities, such as shock waves in supersonic flight or breaking waves in the ocean, presents a significant challenge for high-order numerical methods. Standard approaches often suffer from the Gibbs phenomenon, producing spurious oscillations that can lead to non-physical results or simulation failure.
The ENO procedure, introduced in the late 1980s, mitigates this by employing an adaptive stencil. Rather than using a fixed set of neighboring points, it dynamically selects the "smoothest" possible neighborhood to calculate values. While practically successful, the mathematical community lacked a global, mesh-independent proof of its stability until now. This research fills that 25-year gap by demonstrating that the nonlinear stencil selection process is inherently rigid and stable.
Methodology
The authors analyze two primary versions of the algorithm: ENO reconstruction (recovering a function from cell averages) and ENO interpolation (recovering a function from point values). The proof relies on three core mathematical pillars:
- Newton Representation: Interpolating polynomials are represented using Newton forms, which utilize "divided differences" to measure local function roughness.
- The Sign Property: The authors derived a novel expression for the "jump" at a cell interface, proving it is a sum of terms sharing the same sign as the jump in the original data.
- Recursive Bounding: Using the recursive nature of divided differences, an induction argument establishes that the ratio between the reconstructed jump and the original data jump is always bounded by a constant $C_p$, dependent only on the order of accuracy $p$.
Stability and the Sign Property
The most intuitive finding is the "Sign Property." If the underlying data trends upward at an interface, the ENO reconstruction is guaranteed to follow that direction. It will never create a local "flip" or artificial oscillation that contradicts the trend of the underlying averages. This reveals that the ENO logic is fundamentally robust across any order of accuracy and any non-uniform mesh.
Interestingly, the authors note that popular Weighted ENO (WENO) methods—which use a weighted average of stencils—do not necessarily satisfy this sign property. This suggests that the original "winner-take-all" stencil selection of ENO possesses unique stability benefits.
Key Results
The paper provides concrete upper bounds ($C_p$) for how much the ENO procedure can magnify a jump. On a uniform mesh, these constants are:
- 3rd-order ($p=3$): Jump is at most 3.33x the original data jump.
- 4th-order ($p=4$): Jump is at most 5.33x.
- 6th-order ($p=6$): Jump is at most 13.86x.
Furthermore, the authors prove that if underlying cell averages are equal ($v_{i+1} = v_i$), the ENO reconstruction is guaranteed to be perfectly continuous across that interface. These results serve as the necessary building blocks to prove that ENO-based schemes are "entropy stable," ensuring physical consistency in complex fluid simulations.
Conclusion and Future Work
While this work proves that jumps at interfaces are stable, it does not yet establish Total Variation Boundedness (TVB); it limits the behavior at cell "edges" but does not strictly bound polynomial oscillations within a single cell. Future research aims to utilize these stability results to construct arbitrarily high-order entropy stable schemes for systems of conservation laws, potentially offering a new level of reliability for simulating complex physics like turbulence and combustion.
Summary updated