Introduction
The Advection-Diffusion Equation (ADE) is a fundamental mathematical framework for simulating the transport and spreading of substances—such as pollutants, sediment, or oil spills—within fluid flows. Smoothed Particle Hydrodynamics (SPH), a Lagrangian meshless method, is frequently employed for these simulations due to its ability to handle complex boundaries and follow fluid parcels naturally.
However, standard SPH implementations often struggle with "wiggles"—non-physical numerical oscillations—at sharp concentration jumps or discontinuities. These oscillations can lead to unstable results or physically impossible outcomes, such as negative concentrations. Consequently, researchers often avoid the more mathematically elegant flux-form of the ADE in favor of less robust alternatives. This paper introduces a novel Partial Diffusive Flux (PDF) formulation for SPH that effectively eliminates these oscillations while maintaining high-order accuracy.
Methodology: The SPH-PDF Approach
The core innovation of the SPH-PDF scheme lies in its treatment of second-order derivatives through a specialized flux-splitting technique. The method addresses the mathematical imbalances that occur at discontinuities in standard SPH.
Partial Flux Splitting
Instead of calculating a single global flux for a particle, the algorithm utilizes indicative functions to categorize neighboring particles into two distinct groups: those "in front" and those "behind" the particle relative to the concentration gradient.
Front/Rear Derivatives and Consistency
The method calculates separate "front" and "rear" derivatives. By mathematically correlating these partial components, the PDF method "fills in" the missing terms that typically cause ringing artifacts in standard SPH. To further enhance accuracy, the authors integrated the Corrective Smoothed Particle Method (CSPM). This ensures particle consistency, maintaining kernel approximation accuracy even when particles are irregularly distributed or near boundaries.
Experimental Evaluation
The SPH-PDF approach was validated against five rigorous synthetic benchmarks with known analytical solutions:
- 1D Pure Diffusion Step: Used for theoretical analysis of oscillation causes.
- 1D Discontinuous ADE: A "top-hat" concentration profile undergoing simultaneous advection and diffusion.
- 2D Discontinuous ADE: A circular contaminant patch moving diagonally across a domain.
- 2D Anisotropic Dispersion: A complex scenario where diffusion rates vary by direction, typical of groundwater flow.
- 1D Gaussian Case: Specifically used to test convergence rates and performance under irregular particle distributions.
Key Results and Insights
- Elimination of Oscillations: In both 1D and 2D discontinuous tests, SPH-PDF produced perfectly smooth concentration profiles. In contrast, standard SPH-DF (Diffusive Flux) methods exhibited significant numerical ringing and overshoot.
- Superior Accuracy: The numerical error ($L_2$ norm) of the PDF method was at least one order of magnitude lower than conventional diffusive flux formats.
- Second-Order Convergence: While most numerical methods drop to first-order accuracy at discontinuities, SPH-PDF maintained a convergence rate of approximately 2.0 (2.0193 in 1D and 2.186 in 2D). This means doubling the particle resolution reduces error by a factor of four.
- Anisotropic Performance: In complex anisotropic cases, SPH-PDF reduced negative concentration errors by two orders of magnitude compared to existing state-of-the-art consistent SPH methods.
- Computational Efficiency: The overhead of splitting the flux into partial components is negligible, providing a massive gain in accuracy without a significant increase in simulation time.
Conclusion and Future Work
The SPH-PDF method represents a significant advancement in the stability of Lagrangian fluid simulations. While it drastically reduces negative concentrations in anisotropic diffusion, it does not eliminate them entirely, marking an area for future refinement. Additionally, while the mathematical framework for 3D extension is provided, future work will focus on 3D benchmarking and applying the PDF logic to other fluid dynamics challenges, such as the Laplacian in viscosity or the pressure Poisson equation, to enhance the stability of complex flow simulations like breaking waves.
Summary updated