Several applications are considered in EPEEC: AVBP (Cerfacs, a numerical simulation framework for the study of fluid dynamics and combustion problems), DIOGENeS (Inria, a numerical simulation framework for the study of nanoscale light-matter interaction problems), OSIRIS (INESC-ID, a numerical simulation framework for the study of plasma physics problems), Quantum ESPRESSO (Cineca, a set of numerical tools for the study of electronic properties of materials) and SMURFF (IMEC, a Bayesian matrix factorization framework for building recommender systems with applications to life sciences). The present news focuses on the AVBP application.
The importance of combustion in our society cannot be overstated, as it is present in almost all aspects of everyday life: energy production, transport, industrial processes and comfort. Alternative energies have improved over the years, but optimizing and minimizing the cost and the consequences of combustion remains an important research topic. With this in mind, the AVBP code is used to design and study combustion on a variety of scenarios both in academy and industry.
The AVBP code was initially co-developed by CERFACS and IFPEN to solve the 3D compressible Navier-Stokes equations. Since 1997, AVBP is developed for the simulation of 3D compressible reactive two-phase flows. The code relies on an explicit resolution of the Navier-Stokes equations on hybrid multi-element meshes using the Large-Eddy Simulation approach where the large structures are resolved and small scales are modeled. The LES models available in AVBP are the Dynamic Smagorinsky, WALE and SIGMA models. In order to handle such arbitrary meshes, the data structure of AVBP employs a cell-vertex finite volume formalism.
The numerical method is based on a finite element low-dissipation Taylor-Galerkin discretization in combination with a linear-preserving artificial viscosity model. AVBP was built for massively parallel computations. The code represents approximately 500,000 lines of Fortran with some C elements. Parallelism follows the single program multiple data paradigm and is based on a domain decomposition using the ParMeTiS or Pt-Scotch graph partitioning tools. Large input/output operations are based on the Parallel HDF5 library with the possibility of having a flat (all processes) or hierarchical (one process out of N) I/O pattern depending on the file system contention. On the fly post-processing algorithms allow for data extraction on iso-surfaces and planes to reduce data output. The cell-vertex approach uses cache blocking based on cell grouping to reduce cache errors.
In EPEEC, we aim at removing bottlenecks for more efficient and powerful simulations. The motivation for faster and more reliable exascale computations comes from the multi-scale aspect of turbulence and combustion in fluid dynamics. Gas turbines, rocket engines, piston engines are of the order of the meter in size or more, however flame fronts represent less than a millimeter. Furthermore, these flames can interact with geometric details such as injection tubes and characteristic length-scales of the system triggering unwanted and often destructive instabilities that cost billions of dollars to the industry. More accurate and fine detail simulations are an ideal surrogate for designers to test out ideas at a limited cost.