Discovering physics extremes with computing

This project aims to build theoretical insight and algorithmic tools to explore the dynamic behavior of complex physical systems.

These systems are typically studied via computational modelling, where the equations governing the system’s properties are represented and solved numerically. Complex systems may result in models with billions of degrees of freedom or more, and can only be handled with sophisticated, parallel-computing approaches. Capturing the model’s behavior in many different configurations results in a formidable data-science problem, currently tackled with ad-hoc techniques in different domains. This project aims to introduce a universal framework for identifying special dynamic regimes in large-scale models, representing events that are highly unusual, severe, or otherwise unique. Real-world applications of these techniques span many key sectors, from weather and climate to energy and healthcare.

As these regimes are both rare and complex, a combined data- and knowledge-driven framework is necessary to capture their signatures at scale. The emerging field of physics-informed artificial intelligence, along with classical model-reduction techniques, provide the context for this study.

The project is funded by a UKRI Future Leaders Fellowship (PI: Eloisa Bentivegna, 2021-2025).