This manuscript presents an efficient solver for the linear system that arises from the Hierarchical Poincar\'e-Steklov (HPS) discretization of three dimensional variable coefficient Helmholtz problems. Previous work on the HPS method has tied it with a direct solver. This work is the first efficient iterative solver for the linear system that results from the HPS discretization. The solution technique utilizes GMRES coupled with an exact block-Jacobi preconditioner. The construction of the block-Jacobi preconditioner involves two nested local solves that are accelerated by local homogenization. The local nature of the discretization and preconditioner naturally yield matrix-free application of the linear system. A distributed memory implementation allows the solution technique to tackle problems approximately $50$ wavelengths in each direction requiring more than a billion unknowns to get approximately 7 digits of accuracy in less than an hour. Additional numerical results illustrate the performance of the solution technique.
翻译:此手稿为线性系统提供了一个高效的解答器, 其来源于高级波因卡耳斯泰克洛夫( HPS) 的分解三个维可变系数 Helmholtz 问题。 HPS 方法的先前工作已经与直接解答器捆绑在一起。 这是由 HPS 分解产生的线性系统的第一个高效的迭代解答器。 解决方案技术使用GMRES 和一个精确的组合 Jacobi 先决条件。 块- Jacobi 先决条件器的构建涉及两个嵌套的本地解答器, 由本地同质化加速。 离散性和前提软件自然生成的线性矩阵应用的本地性质。 一个分布式内存实施使解决方案技术能够解决每个方向上大约50美元波长的问题, 需要超过10亿个未知数据才能在不到一小时的时间内获得大约7位的准确度。 其他数字显示解决方案技术的性能 。