SEAVEA is a new project which will develop an exascale-ready toolkit that allows applications across the spectrum of ExCALIBUR to apply VVUQ techniques in a mutually repurposable way.
Computer-based simulations have become a critical component of many fields of scientific research and well beyond. The availability of supercomputers permits studies on topics of major societal importance, including climate change, medicine, and energy production, conducted at ever-increasing scales and scope. However, fundamental principles of scientific investigation must be widely enforced to ensure the reliability of and confidence in the results generated. This hinges on our ability to (i) validate (V) them, normally against experimental observations, but sometimes compared to “gold standard” numerical results; (ii) verify (V) them, to ensure that the correct mathematics is used, and (iii) equip the output data with meaningful bounds on errors which come from rigorous uncertainty quantification (UQ).
Our objectives are to:
- Show how state-of-the-art VVUQ approaches can provide tangible generic benefits at the emerging exascale;
- Further, embed modern data science principles and practice into best practice in VVUQ;
- Establish streamlined orchestration of the various components of multiphysics and multiscale models to optimize the cost of performing VVUQ and the fidelity of the predictions;
- Implement these approaches via a software environment for the UKAEA’s multi-physics nuclear fusion modeling use case;
- Demonstrate the applicability of the same software environment for the Met Office’s weather and climate use case;
- Demonstrate the applicability of the same software environment for the aforementioned DDWG use cases and others through knowledge exchange during the lifetime of this project;
- Disseminate our findings, methods, and software across as wide a range of UK users as possible within Excalibur, the wider academic community, industry, publicly funded research establishments, and international research communities.
Computer simulation results are validated compared with experiment in several ways, ranging from qualitative to quantitative measures which apply a validation metric. Likewise, verification is concerned with confirmation that the mathematical model and corresponding algorithm have been coded correctly. Uncertainty quantification (UQ) is concerned with understanding the origins of and assessing the magnitudes of the errors which accompany computer simulations, whether epistemic or aleatoric.
VVUQ is nec essary for any simulation that makes predictions in advance of an event to become actionable – that is, for its output to be useful in any form of decision-making process, from government interventions in pandemics to the choice of materials to combine for aircraft wing production. Here, exascale computing offers more opportunities to make actionable predictions.
Moreover, because VVUQ is intrinsically compute intensive due to its ensemble-based execution pattern, it too requires exascale resources, as well as advanced resource management strategies to efficiently manage the large numbers of concurrent runs necessary.
We will establish an open source and open development VVUQ toolkit optimised for efficient execution at current pre- and emerging exascale. This will include advanced approaches for surrogate modelling in order to minimise the expense and time needed to perform the most compute-intensive calculations and will demonstrate its efficiency gains for a diverse array of VVUQ workflows within multiple scientific applications, and on architecturally and geographically diverse emerging exascale environments.
The software developed, implemented and benchmarked in this project will become an open and invaluable asset to the UK ExCALIBUR community but also much more widely within UK and internationally as high-performance computing enters the exascale era.
The exascale toolkit will be built on a combination of widely used tools and services which will be evolved to handle systems of increasing levels of complexity. These include components from the VECMA project (EasyVVUQ, FabSim3, QCG-PJ and EasySurrogate), as well as the UCL-Alan Turing Institute Multi-Output Gaussian Process Emulator (MOGP). We will apply these capabilities to several applications, including: (i) the UKAEA’s tokamak fusion modelling use case for which a working software environment will be produced; (ii) weather and climate forecasting for the Met Office; (iii) turbulent flow simulation for environmental science; (iv) prediction of advanced materials properties of graphene-polymer based nanocomposites for aerospace applications; (v) high-fidelity patient-specific virtual human blood flow system for medical research; (vi) drug discovery; and (vii) human migration.
