ABSTRACT The multi-mesh, multi-solver paradigm makes use of multiple Computational Fluid Dynamics (CFD) solvers in a single overset framework. A framework-level implementation of the Generalised Minimal Residual algorithm applied to the full implicit overset system is presented. The method requires only minimal changes to existing, Python-wrapped CFD solvers and demonstrates improved convergence compared to traditional methods. Preliminary validation for the method is performed using the Poisson equation on two overset grids. The multi-body, overset GMRES method is shown to be equivalent to a conventional implementation which allocates memory for data across all overset grids on a single processor. Results are shown for two time-accurate flow simulations. The first case analyses convergence of a 2D inviscid wedge and the second analyses convergence of a laminar 3D sphere. Both the wedge and sphere bodies shed unsteady structures into nested, background Cartesian grids. In both unsteady cases, the presented overset GMRES method demonstrates superior convergence over traditional methods.

中文翻译：

---

多求解器范式的一种超广义最小残差方法

摘要 多网格、多求解器范式在单个重叠框架中使用多个计算流体动力学 (CFD) 求解器。提出了应用于全隐式重叠系统的广义最小残差算法的框架级实现。该方法只需要对现有的、Python 封装的 CFD 求解器进行最小的更改，并且与传统方法相比，收敛性得到了改进。该方法的初步验证是在两个重叠网格上使用泊松方程进行的。多体、重叠 GMRES 方法被证明等效于传统的实现，它在单个处理器上为所有重叠网格中的数据分配内存。显示了两个时间精确的流动模拟的结果。第一种情况分析 2D 无粘楔的收敛性，第二种情况分析层状 3D 球的收敛性。楔形体和球体都将不稳定的结构脱落到嵌套的背景笛卡尔网格中。在这两种非稳态情况下，所提出的重叠 GMRES 方法都表现出优于传统方法的收敛性。