Scientific model algorithms get a makeover to increase confidence in results
Supercomputers run powerful simulations that model some of the greatest challenges scientists are decoding, from achieving nuclear fusion to finding safer drugs and predicting future climate change. The power of accurate prediction requires multiscale calculation of vast amounts of data. But to confirm reliability so researchers have confidence in their results, computer simulations have to both correctly represent processes and accurately quantify the uncertainties associated with their calculations.
The VECMA project (Verified Exascale Computing for Multiscale Applications), funded by the EU Future and Emerging Technologies (FET) programme, aims to raise the bar of confidence in computer simulations results so outputs can be actionable and lead to real outcomes. The project takes simulations – designed to run on current petascale and emerging exascale supercomputers – and runs them through a process which assesses the extent to which these are verified (V), meaning whether the simulations solve the model correctly. The process also includes validated (V), which is whether the simulations’ output agrees with given data. VECMA also quantifies the extent of this agreement or uncertainty (UQ).
During the course of this project, VECMA will design, develop and gradually release a generic toolkit which will allow users to rapidly integrate of VVUQ methods into multiscale applications in a scalable manner. This will be done through generic templates, which can then be customized and aggregated to create a dedicated VVUQ procedure for multiscale applications. The VECMA toolkit will provide support in customising these templates, for example by automatically inserting user- and application-specific information and by automatically creating composite templates, creating a VVUQ-enabled Multiscale Execution Plan. This plan can then either be used to execute the application on existing resources, or to predict the performance of the application on future architectures using an exascale infrastructure.
The goal is to make the toolkit open source and widely available for researchers to use in European high-performance computing (HPC) centres. So the project’s VVUQ algorithms are designed to have high generic applicability to allow easier assessment and certification in a wide array of research areas, enabling important decisions to be based on simulation outputs. In an era where science is afflicted by a ‘reproducibility crisis’ and confidence in findings is questioned, VECMA aims to create a unified European VVUQ package increase confidence that Exascale computer simulations can be trusted tools in the decision-making process for policy makers as well as other researchers.