Introduction
Numerical simulations of the shallow-water equations (SWE) are essential for predicting geophysical phenomena such as tsunamis, storm surges, and ocean currents. A persistent challenge in these models is the treatment of non-flat bottom topography. In a physical "lake at rest" scenario, the water remains perfectly still regardless of the underlying terrain. However, in numerical simulations, the mathematical terms representing the water's pressure gradient and the gravitational source term (the slope of the bed) are often discretized separately. If these discrete approximations do not cancel each other out exactly, the simulation generates "spurious waves"—numerical artifacts where the water begins to move on its own. Schemes that eliminate these errors are known as well-balanced.
While well-balanced methods have been established for finite-volume grids, they are significantly more difficult to design for nodal methods like finite differences and especially for meshless methods. This paper introduces a general mathematical framework to ensure that both traditional grid-based and flexible meshless schemes perfectly preserve the lake-at-rest steady state.
A Mimetic Design Framework
The authors propose a mimetic design strategy, where discrete operators are constructed to mirror the fundamental identities of continuous calculus. The core of the framework is the derivation of a specific algebraic relationship—formalized in the paper's Theorem 2.1—between three discrete components:
- The Flux Derivative Operator: Governs how water and momentum move through the domain.
- The Source Derivative Operator: Represents the influence of bottom topography.
- The Averaging Rule: Determines how water height is weighted at the nodes.
The research demonstrates that a scheme is well-balanced if and only if these three operators are compatible. A key insight of the paper is the "asymmetry in design": it is mathematically simpler to select a source derivative and an averaging rule and then derive the necessary flux derivative than to attempt the reverse. This framework allows for the creation of high-order schemes (including fourth-order) that maintain the well-balanced property.
Application to Meshless Methods
A major contribution of this work is the application of the framework to Radial Basis Function-generated Finite Differences (RBF-FD). Unlike traditional methods that require a rigid grid, RBF-FD allows points (nodes) to be scattered arbitrarily. This flexibility is crucial for modeling complex coastal geometries or using dynamic node adaptation to track a wave as it approaches the shore. The authors prove that their well-balanced conditions hold even when nodes are irregularly distributed, a significant advancement for meshless geophysical modeling.
Experimental Validation
The proposed methods were evaluated using several standard computational fluid dynamics benchmarks:
- 1D and 2D Lake at Rest: These tests involved stationary water over complex topographies (e.g., a "cosine bump"). In both cases, the well-balanced schemes maintained the steady state with errors at machine precision ($10^{-15}$ to $10^{-16}$), effectively zero for 64-bit computing. In contrast, unbalanced schemes produced significant errors in the $10^{-1}$ range.
- Parabolic Bowl: This test simulates dynamic, oscillatory flow (water sloshing). The well-balanced RBF-FD scheme showed significantly improved mass conservation—often an order of magnitude better than standard meshless methods—demonstrating that the framework's benefits extend to dynamic scenarios.
Insights and Limitations
The study reveals that for standard diagonal finite difference schemes, there is a theoretical limit: one cannot exceed second-order accuracy if only two nonzero weights are used in the stencil while maintaining the well-balanced property. To achieve higher-order accuracy, more complex stencils or specific averaging rules must be employed.
While the framework is robust, the authors note that meshless methods like RBF-FD often require "hyperviscosity" (numerical smoothing) to ensure stability, though this does not compromise the well-balanced property. Future work will focus on extending these methods to handle "wetting and drying" (inundation) and optimizing the computational cost of building differentiation matrices for large-scale, real-world applications.
Summary updated