ORB5X: a performance-portable global electromagnetic gyrokinetic PIC code in C++/Kokkos built using Agentic AI
Abstract
ORB5X is a C++17/Kokkos AI-assisted translation of ORB5, a global electromagnetic gyrokinetic particle-in-cell code for toroidal confined plasmas. The translation preserves the mathematical model and the main numerical algorithms of the Fortran implementation, i.e. a Lagrangian gyrokinetic formulation, marker-particle representation of the distribution functions, finite-element B-spline representation of the fields, Fourier filtering in the angular directions, MPI domain decomposition and cloning, and HDF5 diagnostic output. The main software change is the replacement of Fortran modules and allocatable arrays by typed C++ classes, structs, namespaces, and Kokkos Views and kernels for performance portability across multicore CPU and GPU backends. This paper summarizes the physical models, the numerical scheme, the distributed and shared-memory parallel design, the AI-assisted code translation workflow, and the validation path used to compare ORB5X against the original Fortran reference codebase. The initial tests for ITG, ITPA, and chirping show an agreement of almost machine accuracy between ORB5X and ORB5 for the electrostatic and electromagnetic field components. Furthermore, we demonstrate that with the modern CMake and Kokkos, ORB5X can be easily compiled and efficiently run on personal Linux and Mac laptops, as well as on high-performance computing clusters such as LUMI-G, Daint-ALPS, CINECA, and Discoverer.
Related
- Global Gyrokinetic Simulations of Electromagnetic Turbulence in STEP
- Energetic particles, Shafranov shift and finite β effects on TAE, KBM and ITG instabilities in global electromagnetic gyrokinetic simulations
- HEPTABench: A framework for portable and controlled agentic evaluations in high energy physics
- Performance Portable SU(N) Lattice Gauge Theory Simulation with Kokkos
- Performance of a Hot Rb Vapour-Based Portable Quantum Memory
BibTeX
@misc{sadr2026orb5x,
title = {{ORB5X: a performance-portable global electromagnetic gyrokinetic PIC code in C++/Kokkos built using Agentic AI}},
author = {Mohsen Sadr and Emmanuel Lanti and Alexey Mishchenko and Xin Wang and Laurent Villard},
howpublished = {arXiv},
year = {2026},
month = sep,
doi = {10.48550/arxiv.2607.17363},
eprint = {2607.17363},
archivePrefix = {arXiv},
url = {https://arxiv.org/abs/2607.17363},
note = {CC BY 4.0},
}
RIS
TY - UNPB
TI - ORB5X: a performance-portable global electromagnetic gyrokinetic PIC code in C++/Kokkos built using Agentic AI
AU - Mohsen Sadr
AU - Emmanuel Lanti
AU - Alexey Mishchenko
AU - Xin Wang
AU - Laurent Villard
PY - 2026
DA - 2026/09/08
PB - arXiv
DO - 10.48550/arxiv.2607.17363
UR - https://arxiv.org/abs/2607.17363
AB - ORB5X is a C++17/Kokkos AI-assisted translation of ORB5, a global electromagnetic gyrokinetic particle-in-cell code for toroidal confined plasmas. The translation preserves the mathematical model and the main numerical algorithms of the Fortran implementation, i.e. a Lagrangian gyrokinetic formulation, marker-particle representation of the distribution functions, finite-element B-spline representation of the fields, Fourier filtering in the angular directions, MPI domain decomposition and cloning, and HDF5 diagnostic output. The main software change is the replacement of Fortran modules and allocatable arrays by typed C++ classes, structs, namespaces, and Kokkos Views and kernels for performance portability across multicore CPU and GPU backends. This paper summarizes the physical models, the numerical scheme, the distributed and shared-memory parallel design, the AI-assisted code translation workflow, and the validation path used to compare ORB5X against the original Fortran reference codebase. The initial tests for ITG, ITPA, and chirping show an agreement of almost machine accuracy between ORB5X and ORB5 for the electrostatic and electromagnetic field components. Furthermore, we demonstrate that with the modern CMake and Kokkos, ORB5X can be easily compiled and efficiently run on personal Linux and Mac laptops, as well as on high-performance computing clusters such as LUMI-G, Daint-ALPS, CINECA, and Discoverer.
ER -