Interpolation at Arbitrary Points in High-Order Meshes on GPUs

Watch a technical presentation from the 2024 MFEM Workshop exploring robust and scalable arbitrary point interpolation techniques for GPU-based high-order mesh computations. Learn how to efficiently query partial differential equation solutions, compare solutions between meshes, and perform Lagrangian particle tracking using a combination of globally partitioned maps, processor-local maps, and element-wise bounding boxes. Discover how Newton's method with trust region-based approaches can be implemented to invert affine maps and determine reference space coordinates, with specialized kernels developed for GPU acceleration. Examine the effectiveness of these techniques through demonstrations on various high-order meshes, particularly focusing on solutions for large-scale unstructured meshes partitioned in parallel with MPI.