Abstract
We present a novel approach for modeling subduction using a Multipole-accelerated Boundary Element Method (BEM). This approach allows for large-scale modeling with a reduced number of elements and scales linearly with the problem size. For the first time, BEM has been applied to a subduction model in a spherical planet with an upper-lower mantle discontinuity, integrated with a free-surface mesh algorithm.
Introduction
Modeling the dynamics of tectonic plates is a significant multiscale challenge. Large-scale plates spanning 10,000 km interact with narrow lubrication zones only 100 meters thick. Traditional numerical methods, such as Finite Element (FE) or Finite Difference (FD), require volumetric meshing of the entire Earth's mantle. To capture fine-scale details, these methods require an impractical number of elements, making global 3D simulations computationally prohibitive. Furthermore, accurately tracking "free surfaces" where the Earth's crust deforms is difficult in traditional volumetric grids.
Methodology
To address these challenges, we employ the Boundary Element Method (BEM). Unlike volumetric methods, BEM only requires the discretization of interfaces between different materials, significantly reducing the dimensionality of the mesh.
To ensure computational efficiency, we integrate the Fast Multipole Method (FMM). While standard BEM scales as $O(N^2)$, FMM acceleration reduces this to $O(N \log N)$, making large-scale global problems tractable. The Earth's mantle is modeled as a highly viscous fluid using Stokes equations. A key innovation is our adaptive mesh algorithm that adjusts the Earth's free surface to lithospheric topography, ensuring that buoyancy forces—the primary driver of subduction—are calculated self-consistently.
Key Results and Insights
The FMM-BEM approach demonstrates near-linear scaling for high-resolution models, a significant improvement over traditional boundary methods. Performance benchmarks on a Beowulf cluster show that the code maintains approximately 90% parallel efficiency when scaled up to 64 CPUs.
Geologically, the simulations successfully replicate known subduction behaviors. The model highlights how the sharp viscosity increase at the 660km mantle discontinuity affects slab behavior, inducing "mantle back-flow" and influencing trench retreat. The success of the model depends on the "critical distance" ($h_{crit}$) between the surface and the subducting plate; when tuned correctly, the plate is sustained self-consistently without sinking incorrectly.
Limitations and Future Work
Current implementation constraints include memory overhead from the "shared tree" FMM structure, which may limit problem size on specific hardware. Additionally, the model currently uses uniform viscosity layers, whereas real-world lithosphere exhibits complex, non-linear rheology. Future research will focus on investigating more complex plate geometries and the global interactions between multiple subducting plates.
Summary updated