Arguably the most significant scientific challenge in the field of materials simulation is an urgent need to transition from modelling “simple systems in equilibrium” to describing complex quantum phenomena and open systems under “kinetic control”. Use cases exist both in nature and in industrial environments (e.g. construction materials for chemical and nuclear storage/reactors) and Net Zero targeted technologies (e.g. batteries, solar cells, computers and lighting). Larger and longer simulations are needed to describe, for example, exotic quantum properties of materials such as topological insulators that are increasingly being revealed by experiments. Sophisticated theories such as quantum Monte Carlo (QMC) and the random phase approximation are essential to describe accurately the properties of many materials, but to date, current computational capabilities have only allowed limited simulations at these levels.
Examples of key challenges, driven by societal & industrial needs, include:
- The search for novel “smart materials” with self-cleaning and self-healing surfaces and interfaces
- The rational design of target (metastable) materials
- The development of an integrated, in operando representation of active sites in heterogeneous catalytic applications, including non-equilibrium conditions, for accurate quantitative description (and reaction rates) and comparison with experimental data
- Designing nanoscale metamaterials for novel functional devices, e.g. photonic metamaterials for photovoltaic or sensing applications, or phononic metamaterials for thermoelectric applications.
Time-to-solution (strong-scaling) on evolving hardware types (accelerators and other many-core processor-based systems) and handling all the data files produced are also principal challenges.
Activities to date have involved:
- The creation and development of a list of requirements for the community’s software to become “exascale ready”.
- A prioritisation of the list of requirements.
- Review and steering activities undertaken by RSEs and PDRAs with support from the Computational Science Centre for Research Communities (CoSeC).
- Development of blueprint solutions to be used by the above and in Phase 2.
Challenges this working group are addressing include:
Hero-calculations, or stable strong-scaling on heterogeneous architectures – The heterogeneous nature of exascale HPC, coupled with the low-level optimisations required for good performance, mean that it is increasingly difficult for domain scientists to develop high-performance code. Encapsulating the key computational kernels in an application independent manner, with well specified interfaces, is enabling a highly desirable “separation of concerns”.
Complex workflows – Identifying the limitations of the workflow approaches currently employed on petascale HPC systems and the co-design of prototypes that will inform future developments for exascale including; coupling of software packages together to enable complex workflows to be created; scheduling of complex loosely-coupled workflows; development of algorithms that are robust to failure of component calculations.
Exascale Challenge – Data curation, validation, and parallel I/O.