Accuracy and Computational Efficiency of a 3D Viscoelastic Solver with Arbitrary Relaxation Mechanisms Using Devito DSL

Viscoelastic modeling and inversion traditionally use Finite Differences (FD) or Finite Elements (FE) implemented in low-level programming languages like C, which are time-consuming to develop. Recent advancements in Domain Specific Languages (DSLs) offer a transformative approach by allowing geophysicists to write high-level code closely resembling mathematical equations. This code is automatically translated into optimized FD or FE implementations for high-performance computing (HPC) systems, from local machines to HPC clusters. This significantly accelerates development while ensuring robust, efficient performance.

DSL frameworks simplify the complex task of generating low-level, HPC-optimized code, adapting to various architectures such as CPUs and GPUs. Bisbas (2023) previously demonstrated a DSL-based viscoelastic modeling approach with a single relaxation mechanism. Building on this, the current study generalizes the framework to handle an arbitrary number of relaxation mechanisms, producing efficient FD code tailored for HPC environments.

The study examines how the number of relaxation mechanisms affects both computational performance and model accuracy. Results provide practical insights for selecting the optimal model complexity, balancing accuracy with computational demands. This advancement enhances the scalability and precision of viscoelastic modeling, offering a powerful tool for geophysical applications.