TL;DR
This paper introduces a "factored eikonal equation" approach to improve the accuracy of traveltime calculations in wave propagation problems. By decomposing the solution, their method effectively handles singularities and achieves more accurate results than traditional methods, particularly for complex scenarios like point sources.
What Problem Does This Paper Address?
Calculating traveltimes of waves is fundamental in many scientific and engineering fields, including seismic imaging, computer vision, and computational geometry. The eikonal equation is the standard mathematical model for this, but solving it numerically can be challenging, especially when dealing with complex wave speeds or when the solution has singularities (e.g., near a wave source). Traditional numerical methods often struggle to accurately capture these singularities, leading to reduced accuracy in the computed traveltimes. This paper aims to address this accuracy limitation by proposing a novel formulation of the eikonal equation.
Methods
The core innovation is the "factored eikonal equation." Instead of solving the original eikonal equation directly, the authors decompose the traveltime solution $T(x)$ into a product of two factors: $T(x) = T_0(x) \tau(x)$. Here, $T_0(x)$ is a known or easily computable "approximate" solution (like a distance function or a solution to a simpler eikonal equation), and $\tau(x)$ is a correction factor that satisfies a new, "factored" eikonal equation.
The paper then adapts the Fast Sweeping Method (FSM) to solve this factored equation. FSM is an efficient iterative technique that uses Gauss-Seidel iterations with alternating sweep directions to solve Hamilton-Jacobi equations. The key to FSM's efficiency is its ability to propagate information along characteristics in a finite number of iterations, independent of the grid size. The authors design a specific discretization and update strategy for the factored equation, ensuring that the causality of the original eikonal equation is preserved during the iterative process.
Datasets
Not discussed in the paper. The paper focuses on numerical experiments using analytical solutions and a well-known benchmark model (Marmousi).
Key Insights and Ideas
- Factoring for Accuracy: Decomposing the traveltime solution into a known approximate part ($T_0$) and a correction part ($\tau$) is a powerful way to isolate and handle singularities. $T_0$ can be chosen to capture the singular behavior (e.g., distance from a source), leaving $\tau$ to be a smoother function that is easier to approximate accurately.
- FSM Adaptation: The Fast Sweeping Method, known for its efficiency in solving eikonal equations, can be effectively adapted to the factored form. The key is to ensure that the FSM's update scheme respects the causality of the original eikonal equation, even when applied to the factored form.
- Improved Convergence Behavior: The paper claims that the convergence behavior of the FSM for the factored eikonal equation is similar to that of the original eikonal equation, meaning the number of iterations is independent of mesh size. This preserves the computational efficiency of FSM.
- Singularity Handling: The primary benefit of the factored approach is its superior handling of singularities, particularly those arising from point sources. This leads to significantly more accurate traveltime solutions compared to directly solving the original eikonal equation.
Key Results
- Accuracy Improvement: Numerical experiments consistently show that the factored eikonal equation approach yields significantly more accurate traveltime solutions than the direct FSM application to the original eikonal equation. For example, in Table 1, the error for the factored method is orders of magnitude smaller than for the original method.
- Efficiency Maintained: The FSM applied to the factored equation maintains the efficiency of the standard FSM, with the number of iterations remaining constant and independent of mesh size across various examples.
- Marmousi Model Performance: When applied to the complex Marmousi model, the factored eikonal equation method achieved better accuracy and required fewer iterations (18 vs. 28) than the original FSM, though with a slightly higher CPU time (93 sec vs. 62 sec).
Limitations and Future Work
- Choice of $T_0$: The effectiveness of the factored approach relies on a good choice for the approximate solution $T_0$. While distance functions are shown to be effective for point sources, the paper suggests that other analytical or numerical approximations could be used. The paper explores different choices for $T_0$ in its examples, demonstrating how different choices impact accuracy.
- Generalization: While the paper focuses on 2D and mentions extension to higher dimensions, further exploration of the method's performance in 3D and for more complex wave phenomena might be beneficial.
- Computational Cost: While the number of iterations is independent of mesh size, the computational cost per iteration might be slightly higher for the factored equation due to the more complex update rule. The Marmousi example shows a higher CPU time for the factored method, though this is attributed to the specific implementation and tolerance. Future work could focus on optimizing the computational cost.
Summary updated