ESCAPE-2‎ > ‎Project impact‎ > ‎

European excellence in mathematics and algorithms for extreme parallelism and extreme data applications to boost research and innovation in scientific areas such as physics, chemistry, biology, life sciences, materials, climate, geosciences, etc.

ESCAPE-2 conducts European frontier research in algorithm development, by pooling the best European expertise and international partners in large-time-step advection algorithms and highly-scalable, communication-minimizing discretizations with very large domain decompositions. This work provides a significant stepping stone towards the use of time-parallel algorithms, which today are not sufficiently mature for any practical applications in weather and climate prediction. ESCAPE-2 is designed to overcome the current barrier that exists for algorithms in weather and climate that are highly scalable, but inefficient in both time- and energy-to-solution. This work has the potential to substantially influence the future European research agenda for geosciences and beyond (e.g. such technology would readily apply to other potentially time-critical applications such as seismology or biomedical applications where DG-approaches are considered).

With ESCAPE-2, today's proven and fastest algorithms for numerical weather prediction and climate, including both ocean and atmosphere components, will be made portable and optimisable for emerging HPC via an easy-to-learn front-end to a domain-specific language toolchain. This approach will formalise an interface between HPC optimisation/porting experts and novel mathematical algorithm developers beyond the community through the separation of concerns.

ESCAPE-2 will introduce the concept of surrogate neural network models into time-critical applications. By essentially moving extensive training periods – utilising the world’s largest meteorological archive as well as specialised expert models (on radiation) – outside the time-critical path in the workflow, and subsequently transforming low-flop operations typical in physical parametrizations with efficient matrix-multiply operations that apply the training period results, substantial accelerations of specific key components in Earth system modelling are expected.

Lastly, with weather and climate prediction being at the forefront of highly important, highly scalable and immensely time-critical computer applications running routinely on up to O(105 - 106) computer cores, ESCAPE-2 will explicitly address real-world, time-critical application resilience. Through suitably modifying existing mathematical algorithms as well as developing novel mathematical concepts, ESCAPE-2 will substantially enhance European excellence in critical application resilience.

The four themes combine to substantially leverage developments on the route to exascale by overcoming key transition barriers, and in so doing ESCAPE-2 will also facilitate continued European excellence in mathematical and algorithmic developments.