Overview
A unified sparse background-grid framework has been developed for large-scale simulations utilizing the Material Point Method (MPM). This framework aims to address inefficiencies encountered in standard MPM when material occupies a limited portion of the computational domain, a common characteristic of large-scale geophysical mass flow problems. The approach defines sparse grid construction as a general active-node indexing problem. Implementations tailored for different architectures, specifically a scan-based strategy for CPUs and a hash-based strategy for GPUs, realize this sparse framework.
Research Context
The Material Point Method (MPM) operates as a hybrid particle-grid method, frequently employed for analyzing large deformation problems that exhibit history-dependent behavior. Geophysical mass flows represent one such application area where MPM is widely used. Conventional MPM implementations typically rely on a dense background grid. This dense grid can lead to considerable inefficiencies when the material being simulated occupies only a small fraction of the total computational domain. This condition, characterized by sparsity, is often present in many large-scale geophysical mass flow scenarios.
Approach
The research introduces a unified sparse background-grid framework designed for large-scale MPM simulations. The core of this framework lies in treating sparse grid construction as a general active-node indexing problem. To implement this framework across different computing architectures, two distinct strategies were developed:
- A scan-based strategy, optimized for Central Processing Units (CPUs).
- A hash-based strategy, designed for Graphics Processing Units (GPUs).
The proposed framework was evaluated using both benchmark problems and a large-scale landslide simulation.
Findings
Through testing, the unified sparse framework demonstrated several key findings:
- The framework produces results identical to those obtained using standard dense MPM.
- In cases exhibiting strong sparsity, the framework reduced computational time by one to two orders of magnitude.
- In strongly sparse cases, memory usage was also reduced by one to two orders of magnitude.
These reductions in computational time and memory usage were observed in both benchmark problems and a large-scale landslide simulation.
Why This Matters
The described framework addresses a efficiency limitation of standard MPM by enabling more efficient simulations for problems where simulated material occupies only a small portion of the computational domain. This is particularly relevant for large-scale geophysical mass flow problems.