针对多层网格格子 Boltzmann 方法(lattice Boltzmann method, LBM), 发展了一种无需时间插值的、基于单层缓冲区的粗细网格信息传递优化算法. 该算法利用四叉树结构, 构建不同层级格点之间的父子关系, 简化了粗细网格交界面上的分布函数转换, 可以充分发挥LBM 的局部性与并行性. 基于该算法开发了一套通用的多层网格 LBM 并行程序, 对顶盖驱动方腔流、点声源扩散、NACA0012 翼型绕流问题进行了数值计算, 计算结果验证了该算法和程序的有效性及其良好的加速比、效率和可扩展性等并行性能. 与标准 LBM 相比, 本文基于单层缓冲区的多层网格 LBM 在计算稳定性、收敛速度和计算效率等方面具有明显优势.

Based on the multilayer-mesh lattice Boltzmann method (LBM), an information transfer optimization algorithm using a single-layer buffer without time interpolation is developed in this study. The algorithm uses a quadtree structure to construct the parent–child relationship between lattice points at different neighboring levels of mesh, simplifies the transformation of distribution function on the interface of coarse and fine meshes, and fully exploits the locality and parallelism of LBM. A general parallel program of the multilayer-mesh LBM is developed to simulate lid-driven cavity flow, point sound source diffusion, and flow around an NACA0012 airfoil. Numerical results show that the developed algorithm and program are very effective and exhibit excellent speedup, efficiency, and scalability. Compared with the standard LBM, the multilayer-mesh LBM with a single-layer buffer has obvious advantages in terms of computational stability, convergence speed, and computational efficiency.