Publication Date: 2025
Journal of Peridynamics and Nonlocal Modeling (2522896X)7(4)
Peridynamics (PD) provides a robust framework for fracture mechanics but is often limited by high computational costs, particularly in large-scale simulations. The collocation peridynamic scheme, which approximates the PD integral using a set of collocation points, offers a promising approach to reduce this expense. Since this scheme is only applicable to nodes with a complete neighborhood, it is combined with the standard PD scheme to form a general, high-performance hybrid approach. This study presents a comprehensive performance analysis of the hybrid scheme, comparing shared-memory parallelization on multi-core CPUs using OpenMP with a massively parallel GPU implementation using CUDA. During the development of these high-performance solvers, we identified and resolved a critical numerical instability arising in the collocation formulation on very fine grids. Our results show that the GPU-based CUDA implementation achieves substantial speedups (up to 44x) over the serial baseline and significantly outperforms the optimized OpenMP version. This work provides not only a highly efficient computational framework but also improves the reliability of the hybrid scheme, enhancing its applicability for high-fidelity fracture and wave propagation simulations. © The Author(s), under exclusive licence to Springer Nature Switzerland AG 2025.
