To develop and validate a parallel multiphysics suite of solver able to be scalable at the exascale.

Many cutting-edge coupled fluid- structure simulations will require exascale applications. A typical case would be the LES simulation of flow past a wing. Accuracy requirements for LES simulation mean 1013 gridpoints and 107 timesteps for the fluid, which makes exascale computing a promising solution to reproduce realistic scenarios. Moreover, fluid-structure (and even thermal for highspeed flows) coupling will be required. As part of this task, we will develop algorithms and techniques to scale the loose coupling techniques developed over the last 15 years into the exascale range.

Differently to the previous WP, to efficiently run a multiphysic analysis code on parallel architectures, other questions arise than those when scaling a single simulation code. For example, the large range of scales, the syncrhronization of fields within the coupling schemes, the load balancing, combination of structured grids with unstructured points or meshes, and so on.

Specific areas of research and development in this task will include: scalable search algorithms; general, fast, scalable information transfer (fluid, structure, thermal ...) and scalable projection schemes (for conservative force/flux transfer), among others.

As a starting point, we will use the collection of FSI algorithms and codes developed at CIMNE and implemented in KRATOS. These codes have already been ported to an OpenMP/shared memory environment, thus facilitating the realization of this task.

Collection of algorithms for implicit dynamics solvers machines towards the exascale, including description of commonality developments.

Task leader: **CIMNE**. Partners involved: **CIMNE, LUH-IKM, NTUA, QUANTECH**

The simulation of thermo-mechanical problems represents an application for which different levels of accuracy can be used. While engineering accuracy is reachable for many problems of interest at relatively low computational cost, detailed simulation is extremely computationally intensive. A good example is the simulation of the crystallization process during the cooling of metal alloys in casting operations. Since this has a crucial impact on the properties of the final alloy, it is interesting to follow the crystal growth as the size of crystals changes from the micro to the mesoscale. The simulation of such effect for industrial pieces would imply strict requirements on the time and mesh discretization thus requiring an immense computational cost. In this task we will extend the implicit FEM solvers for fluid and solid mechanics developed in WP4 for simulation of large scale thermal-mechanical problems.

The use of these solvers in the next-generation hardware would thus provide an enabling technology to assess the material performance for critical metal pieces which are manufactured via complex thermal-mechanical processes.

Collection of algorithms for implicit dynamics solvers machines towards the exascale, including description of commonality developments.

Task leader: **LUH-IKM**. Partners involved: **CIMNE, LUH-IKM, NTUA, QUANTECH**

The physical phenomena taking place in fusion technology are extremely complex. They involve length and time scales of very different orders, leading to truly multi-scale problems. Implicit algorithms are the way to obtain meaningful large scale results without the need to solve for the smallest scales due to stability constraints. However, they require solving a (non-)linear system at every time step, and so, scalable linear solvers for massively parallel computations are needed.

In this task we will develop FEM fluid models for analysis of plasmas in Tokamaks. This kind of problems are governed by a set of MHD compressible flow equations (the so called Vlasov-Maxwell equations). Fluid-based models do not have the quality of kinetic models, but they are more than enough for many technological problems in fusion reaction. We remark that plasma simulation is possibly the most important application that does require exascale machines. A deeper understanding of plasmas will only be possible by the combination of extremely scalable plasma solvers, as those aimed in this task, and exascale computers.

In this task we also aim to develop incompressible (inductionless) MHD solvers, necessary for the simulation of blanket modules in the fusion reactor, where liquid metals are electrically conducting fluids governed by the MHD equations. In these, cases fluid flows are incompressible, and in operational regimes, the inductionless approximation can be considered.

Collection of algorithms for implicit dynamics solvers machines towards the exascale, including description of commonality developments.

Task leader: **CIMNE**. Partners involved: **CIMNE, LUH-IKM, NTUA**

Lead beneficiary: **CIMNE**

Lead beneficiary: **LUH**

Lead beneficiary: **CIMNE**

Thursday, July 9, 2020
[ login ]

**
611636
**