Skip to main content

Materials design at the eXascale

Periodic Reporting for period 2 - MaX (Materials design at the eXascale)

Reporting period: 2017-03-01 to 2018-08-31

Today the search and discovery of new materials, and the understanding and design of their properties, is a very long and expensive process, mostly based on experimental studies. A key idea behind MaX is that this process could be accelerated efficiently searching for new materials through computer simulations. This would bring down the time-to-market, cut costs, and allow radically new strategies. The increasing accuracy and predictive power of simulations, combined with increasing performance, data capacity and energy sustainability of High-Performance Computing (HPC) technologies, enable a paradigm shift in materials design and discovery: increasingly complex materials behaviour will be addressed by easily accessible and easy-to-use computational experiments.

MaX is a user-driven European Centre of Excellence supporting developers and end-users of materials simulations, design and discovery. It focuses on enabling the best use and evolution of current HPC technologies by creating an open ecosystem of knowledge, capabilities, open software applications, data workflows, analytic tools and user-oriented services.

At the same time, MaX research is enabling the so-called ‘exascale transition’ in the materials domain, by producing the advanced software that will allow the best use of the ‘exascale hardware’ that will be available in a handful of years from now.

This project started in September 2015 and ended in August 2018. The MaX Centre of Excellence will continue in other forms: see
* Flagship codes. MaX centers around a few flagship codes, among the most used and successful in materials research worldwide, that base their predictivity and accuracy on a full quantum mechanical description of the material constituents. MaX has rationalized their design into three distinct levels: math and domain specific libraries (DSL), ‘quantum engines’, and calculation of the desired materials properties. Within this scheme, independent program units now communicate via well designed and documented application program interfaces. Performance and portability of the codes are enhanced by optimizing mathematical libraries and extending the concept to DSL, addressing common tasks among codes. The improved software architecture boosts development, facilitates maintenance, and simplifies interfaces with other codes. This helps interactions with end-users and independent software vendors: e.g. the open-source MaX code Quantum ESPRESSO is now also integrated into a commercial software, increasing its impact on industry. The number of citations of MaX codes has now reached more than 2500 per year.

* Exascale-oriented work. To prepare for the exascale transition, MaX actions are threefold: i) refactoring the codes with different programming paradigms, ii) developing low level code-independent frameworks, to concentrate efforts on a common set of domain specific libraries; iii) implementing software-hardware co-design processes. As a result, MaX codes are considered now a benchmark by all top hardware companies, which include them into their development cycle; moreover, the impact of MaX codes towards green computing is now tangible as the first data in Europe on ‘energy to solution’ are now for the first time available for MaX flagship codes.

* Workflows, data management, high-throughput computing. At the basis of MaX workflows we have an innovative materials informatics infrastructure, Aiida, through a close collaboration with the Swiss project ‘Marvel’. Plugins were developed for all the MaX flagship codes. Workflows were also developed to produce automated benchmarking, crucial for the exascale transition and co-design. The same technology is exploited for data management, data analytics and data sharing, as well as for automatic protocols allowing to launch calculations on different HPC centers, a crucial step towards federation strategies. High throughput research is heavily relying on this technology. Examples of relevant impact are in the study of 2D materials exfoliation and in the design of solid state batteries.

* Serving end-users in industry and research. We facilitate use, data management and output control of codes by developing turn-key solutions based on the above-mentioned workflows.
In addition, the MaX Users Portal delivers services from code download to help-desk, advanced support, and consulting. Quantum-as-a-Service, a user-centered, easy-to-deploy and easy-to-execute VM-like service, allows effective transfer to SMEs industrial users. The MaX observatory has unfolded an “industry wish-list” from the dialogue with industrial end-users and pilot activities with four MaX “lead industrial users”, and developed best practices and workflows that will be made available also to other companies, especially SMEs.

* Training and education. MaX has invested a great effort in training: schools, workshops, contributions to Master and PhD courses, training through research in MaX laboratories received excellent evaluations from participants. In this way we contribute to the current needs of the scientific community and support the preparation of skilled workforce for the future.
Thanks to the collaboration between code developers, experts from HPC centers, and advanced users in the materials domain, MaX progressed well beyond the state of the art in HPC: e.g. we now run multi-petaFlop real-life calculations (not just test routines, but complete workflows yielding results of scientific relevance), and are getting closer to the exascale target in many respects. It is fair to say that Europe is at the forefront of global developments in HPC applications in the field of materials, and MaX is contributing significantly to this achievement. As such, MaX is positioned as a key player in the future of European HPC as a whole.

MaX codes have become a crucial instrument also for HPC/software co-design in the EuroHPC. Some of the MaX codes are now routinely used as a benchmark by hardware companies in the design process. The knowledge emerging from this process is in turn transferred into the development of the codes, in a co-design, self-reinforcing process, in close collaboration with the whole EuroHPC ecosystem. The combination of these three issues - codes with their workflows, services and co-design – represents the knowledge base of MaX for the impact and the future of the CoE.