Organization: Project organization

Level 1 (bottom level): towards generic and scalable numerical algorithms. At the bottom level, the expected contributions will be concerned with:

• Computer science topics which will be in the spirit of this project upstream from the core topics which are centered on the development of high performance numerical schemes and algorithms, and that will deal with some issues that are crucial for exploiting the processing capabilities and raw performances of massively parallel heterogeneous systems.
• Algorithmic aspects in the broad sense with emphasis on the development of generic numerical libraries and solvers in order to benefit from all the parallelism levels with the main goal of optimal scaling on very large numbers of computing entities, taking into account resilience issues.
• Robustness, accuracy and scalability issues of numerical schemes independently of a particular physical context (i.e. generic design issues of high performance numerical schemes for systems of partial differential equations).

Associated activities will be undertaken by researchers from INRIA project-teams that will define the core of the project partnership, gathering:

• Computer scientists whose research and development activities aim at exploiting the processing capabilities and raw performances of massively parallel systems.
• Algorithmists that propose algorithms and contribute to generic libraries and core solvers in order to take benefit from all the parallelism levels with the main goal of optimal scaling on very large numbers of computing entities.
• Numerical mathematicians that are studying numerical schemes and scalable solvers for systems of partial differential equations in view of large-scale numerical simulation of complex physical phenomena.

Level 2 (intermediate level): towards robust, accurate and highly scalable numerical schemes for complex physical problems.

At the intermediate level, the activities will be concerned with the study of the systems of partial differential equations that model the scientific and engineering use cases considered in the project. Topics of interest include 9 discretization in space of underlying systems of partial differential equations (high order approximation, adaptivity, etc.), solution algorithms based on continuous models (domain decomposition algorithms, physics based preconditioners, etc.) and numerical methods adapted to multiscale and multiphysics problems.

These activities will be undertaken by the numerical mathematicians of the core project-teams and from researchers of associated INRIA project-teams that are experts of the underlying physical models and potential users of HPC facilities (these associated teams do not consider HPC as a central topic of their activities and typically rely on the methodologies and tools proposed by the researchers form the core project-teams).

Level 3 (top level): towards exascale computing for the simulation of frontier problems. At the top level, the activities will be concerned with the deployment of large-scale simulations using high performance numerical computing methodologies resulting from the activities undertaken in the bottom and intermediate levels. Numerical results will be validated through comparisons with existing simulation tools in the user community (when this will be possible) or with results published in reference papers of the considered application domains. The success of these demonstration activities will greatly rely on the involvement of external partners from research laboratories or industrial groups that will help in defining and dimensioning a number of frontier problems.