Adaptive Mesh Refine

[xinxiangsama]

Future Plans for Adaptive Mesh Refinement (AMR) Support?

Hi Cabana Team,

First off, thank you for developing and maintaining Cabana. It's a powerful and impressive library for particle-based simulations on modern hardware.

I have a question regarding the future roadmap for the library. I am particularly interested in Adaptive Mesh Refinement (AMR) capabilities. For many large-scale multiphysics simulations, especially those involving shocks, cracks, or complex fluid dynamics, the ability to dynamically refine the grid in specific regions of interest is crucial for both accuracy and computational efficiency.

I was wondering if you could share any information on whether implementing AMR support is part of the future development plan for Cabana or its ecosystem (perhaps through integration with other libraries like ExaAM).

Specifically:

• Is native AMR support a planned feature for a future release of Cabana?
• Are there any ongoing efforts or experimental branches related to AMR that the community could look at?
• What would be the recommended approach for users who currently need AMR-like functionality with Cabana-based applications?

Any insights you could provide on this topic would be greatly appreciated. It would be very helpful for planning future research projects and application development.

Thank you for your time and all the great work!

Best regards,

[xinxiangsama]

I know a lib named "AMReX" can do this,but i think cabana is a more tiny and elegant lib.

[JStewart28]

Hi @xinxiangsama, I am currently working on a small AMR library using Cabana for use in my group's code Beatnik. Please let me know if you talk more about my work and how it could fit into your future research.

[streeve]

Thanks for your interest in Cabana! We use grids primarily to support the particle dynamics, so grid load balancing based on particle populations or the use of sparse grids to avoid grid allocations where there are no particles is as close as we currently get.

We don't have any immediate plans for AMR development internally, but @JStewart28 's efforts would be my suggestion if that's needed for your Cabana-based AMR.

For the record, AMReX is a very well developed library and is entirely focused on AMR. Our "particle first" approach may or may not be useful for your needs

[xinxiangsama]

Hi @xinxiangsama, I am currently working on a small AMR library using Cabana for use in my group's code Beatnik. Please let me know if you talk more about my work and how it could fit into your future research.

hi @JStewart28 , such a work is so great.I'm building a rarefied gas simulator based on direct simulation monte carlo methold,using cabana. For the Sparta(a well knownd dsmc solver),it can refine or coarse gird cell based on cell particle num or macro gird var, it's quite important for a efficient methold for dsmc.I'm very interested in your amr work."xinxiangsama@shu.edu.cn" is my email , can we contact further?

[JStewart28]

Hi xinxiangsama,

I sent you an email but it didn't go through, so I'll respond here.

This is the library, named Tessera, that I am developing:

https://github.com/JStewart28/Tessera

Tessera follows a similar design framework to Cabana. In essence, I use Cabana::AoSoAs to store connectivity data for a triangular unstructured mesh composed of vertices, edges, and faces. (See src/Tessera_Mesh.hpp) Then like Cabana, you can create Arrays attached to vertices, edges, or faces to store data and iterate over (see src/Tessera_Array.hpp) and halo those arrays (See src/Tessera_Halo.hpp).

The build_sphere example uses the VTK library to read vertex, edge, and face connectivity data for a uniform sphere, created by generate_sphere_mesh.py, and perform refinement on the faces.

Another example of Tessera is in Beatnik (the rising-bubble branch) with examples/risingbubble.cpp as the driver problem. Beatnik is being developed to support a finite difference method using Tessera for unstructured meshes. The file src/VTKWriter.hpp writes the unstructured mesh to VTU files that can be visualized with ParaView. Eventually the code in VTKWriter.hpp will be moved into Tessera.

The library needs a fair bit of cleanup and documentation and is far from production ready. But hopefully it gives you an idea of how Cabana can be used to build an adaptive mesh.