A high-performance and portable asymptotic preserving radiation hydrodynamics code with the M1 model

Aims. We present a new radiation hydrodynamics code, called "ARK-RT" which uses a two-moment model with the M1 closure relation for radiative transfer. This code aims at being ready for high-performance computing, on exascale architectures. Methods. The two-moment model is solved using a finite volume scheme. The scheme is asymptotic preserving to capture accurately both optically thick and thin regimes. We also propose a well-balanced discretization of the radiative flux source term able to capture constant flux steady states with discontinuities in opacity. We use the library Trilinos for linear algebra and the package Kokkos allows us to reach high-performance computing and portability across different architectures, such as multi-core, many-core, and GP-GPU. Results. ARK-RT is able to reproduce standard tests in both free-streaming and diffusive limits, including purely radiative tests and radiation hydrodynamics ones. Using a time-implicit solver is profitable as soon as the time step given by the hydrodynamics is 50-100 times larger than the explicit time step for radiative transfer, depending on the preconditioner and the architecture. Albeit more work is needed to ensure stability in all circumstances. Using ARK-RT, we study the propagation of an ionization front in convective dense cores. We show that the ionization front is strongly stable against perturbations even with destabilizing convective motions. As a result, the presence of instabilities should be interpreted with caution. Overall, ARK-RT is well-suited to study many astrophysical problems involving convection and radiative transfer such as the dynamics of H ii regions in massive pre-stellar dense cores and future applications could include planetary atmospheres.