DIAPHANE: A common platform for application-independent Radiative Transport in astrophysical simulations

PI: Lucio Mayer (University of Zurich)

Co-PIs: Matthias Liebendoerfer, Romain Teyssier, Martin Gander, Claudio Gheller, Thomas Peters, Ruben Cabezon, George Lake, Ben Moore, Friederik Karl Thielemann, James Wadsley

January 1, 2014 - December 31, 2016

Project Summary

Astrophysical simulations are evolving along two parallel paths. The domain size and dynamic range of individual simulations is becoming increasingly large, owing to exploitation of massively parallel supercomputers. Likewise, the physics implemented the various types of numerical codes available to the community, grid-based as well as particle-based, is becoming increasingly rich, as demanded by the ever more complex questions that researchers want to address. Increasing the size of the simulation as well as its physical complexity requires that the extra physics modules apply to a wide range of regimes and are flexible and efficient enough to avoid increasing prohibitively the computational cost. The inclusion of radiative physics in simulations across various astrophysical areas, from star and planet formation to galaxy formation, from supernovae modeling to black holes, represents the current most severe bottleneck. Individual codes include only very limited approximations to the radiative transfer equation and to the Boltzmann equation for neutrino transport, which work only in very specific regimes, have implementations strictly tailored to individual codes and are often not the most efficient choice for the regime of interest. Yet it is a recurrent finding that key results depend strongly on the adopted approximation to radiative transfer; gas giant planets may form or not by fragmentation in a protoplanetary disk depending on the underlying thermodynamical evolution, black holes at the center of galaxies remain tiny or become gigantic depending on the accretion regime which is controlled by radiation physics, galaxies might become ionized deeply in their cores or only in their envelopes during cosmic reionization depending once again on the adopted approximation for radiative transfer. Furthermore, certain new problems, such as black hole formation via direct gas collapse, require an efficient implementation of both radiation and neutrino transport in 3D gravitational collapse simulations, which is currently not available. In order to make significant progress in all these fundamental areas of research we need to be able to adopt different approximations to radiative and neutrino transfer in the same simulation. Ideally one wants to be able to switch between different regimes in a single calculation, checking the effect of different approximations and eventually adopting the most efficient approximation for each regime. The long term goal is to achieve "adaptive algorithms" for radiation hydrodynamics, leading to a quantum leap forward in the same way as the emergence of adaptive grids has been a fundamental step forward in the past. The first step in this direction is the establishment of a library of radiation and neutrino transport modules that can be interfaced to any of our community codes, grid-based as well as SPH. This is the main goal of this proposal. The swiss astrophysical community has already developed several modules which encompass most of the relevant approximations We aim at making the modules portable and flexible, with standardized interfaces that can make them exchangeable between the different codes. The second goal is to be able to couple robustly and efficiently the different underlying approximations. Following the pathway already taken for neutrino transport in the modeling of supernovae explosions, we will adopt the Fuzzy Domain Decomposition Method to couple different approximations. This will allow to switch efficiently between them within individual simulations, laying down the path to adaptive radiation hydrodynamics. With the library and the coupling algorithm in place we will be able to readily tackle new applications, some of which we expect to explore already during our co-design project. By standardizing the modules the library will facilitate the optimization of time-critical modules for the new emerging HPC facilities such as massively parallel hybrid GPU-CPU clusters. Optimization of the library for new hardware will begin in due course and will be the main target of a follow-up co-design project. Our project should thus be viewed as a milestone in the development of astrophysical simulations for HPC facilities. We have assembled a diverse team with strong and complimentary expertise in all the key areas. The Zurich group, with PI Mayer and Co-Is Teyssier, Peters, Lake and Moore, has a strong expertise in radiation hydrodynamics simulations in most areas of structure formation, from planets to cosmology, carried out with both AMR and SPH codes (RAMSES, GASOLINE/ChaNGa), some of them already capable to exploit GPUs. They have a strong expertise on massively parallel simulations on the largest supercomputing facilities worldwide. The Basel group, with main Co-I Liebendrfer, Thielemann and Cabezon, is world leader in the modeling of supernovae explosions with state-of-the-art grid-based and particle-based codes (ELEPHANT, SPHYNX), adopting various schemes for neutrino transport. The Geneva group, with Gander, is leader in FDDM methods and has already applied them to the Boltzmann equation in collaboration with the Basel group. Gheller at CSCS/ETH is an expert of high performance computing, including programming for GPU hardware, and has developed state-of-the-art visualization tools (SPLOTCH) adopting raytracing algorithms that can be employed also for radiation hydrodymamics simulations.