While the experience of turbulence is all around us, the mathematics behind remains a mystery. So much so it's one of seven Millennium Prize Problems posed by the Clay Mathematics Institute. These problems challenge the field of mathematics to solve some of the "deepest, most difficult problems" of classical physics.

To improve the accuracy of turbulent flow calculations, Peter Vincent and his team ran thousands of turbulent flow simulations, each requiring billions of calculations to complete, over a period of 12 months. To power these, the team made use of two of Europe's fastest supercomputers - Piz Daint at CSCS and Wilkes-2 from the University of Cambridge.

These NVIDIA GPU-accelerated systems enabled the team to identify for the first time so-called "eigenmode" solutions of averaged turbulent flow in a channel. This provides fundamental insights into the flow physics, which can be used to develop improved approximate models for use in industry.

"From these calculations, we've been able to shed new light on the physics that governs averaged properties of turbulent flow," explained Vincent. "In particular, they show that the governing equations cannot possess certain symmetries, which are often assumed by existing models."

More Information: blogs.nvidia.com