Architecture Advice: Best setup for multi-patch/multi-block systems with local coordinates

[lwJi]

Hi everyone, we are currently designing a multi-patch system for GRMHD simulations. Our primary requirement is that each patch must maintain its own local coordinate system.

We want to ensure we are swimming with the current of AMReX's design rather than fighting it. We'd appreciate your advice on the best architectural setup.

Context & Requirements:

• Grid topology: The patches will be something like cubed-sphere but can be more complicated.
• Data exchange: We will need to interpolate/transfer data across the patch boundaries.
• AMR: We do plan to use Adaptive Mesh Refinement within the individual patches.

Questions:

1. Data Structures: Is it better to instantiate separate AmrCore (or AmrMesh) instances for each patch, or handle this via a single instance using custom Geometry mapping?
2. Boundary Communication: What is the recommended way to handle ghost cell exchanges across these disparate coordinate systems? Is there an existing AMReX utility for inter-grid interpolation, or do we need to write custom MPI packing/unpacking routines?
3. Prior Art: Are there specific tests, tutorials, or external projects using AMReX that already implement a multi-block or mapped-grid topology that we could study?

Thanks in advance for the guidance!

[WeiqunZhang]

You can look at Tests/MultiBlock/Advection for an example of using multiple AmrCore's. It also shows how you can exchange data between blocks. As for, AMR, if the coarse/fine boundary crosses the block boundaries, it might not be easy to do.

[lwJi]

Thanks @WeiqunZhang, I really appreciate it