Yale is partnering with seven other U.S. research institutions to tackle one of the most frenetic phenomena in physics, and the universe — turbulence.
On Aug. 26, the National Science Foundation announced a $30 million grant to fund TEMPEST (Transformative Explorations in Multi-Physics and Engineering of Scientific Turbulence), a science and technology center focused on understanding and controlling turbulence.
TEMPEST, which will be based at Michigan State University, brings together physicists, mathematicians, engineers, and AI researchers — including Yale’s Anna Gilbert, the John C. Malone Professor of Statistics and Data Science in the Faculty of Arts and Sciences, and Rajit Manohar, the John C. Malone Professor of Electrical & Computer Engineering and Computer Science in Yale Engineering.
Anna Gilbert and Rajit Manohar
Turbulence is one of the oldest unsolved problems in physics and a century of work has made only the textbook cases tractable: taking one ordinary fluid, a simple geometry, and one physical effect at a time, with a clean separation between the large scales where energy enters and the small scales where it dissipates. In that setting, turbulence has universal statistics, and a simulation can substitute a generic rule for the scales it cannot resolve.
Turbulence in the real world is a different story.
“Unfortunately, the turbulence inside a fusion reactor, around a reentering spacecraft, or in a dying star obeys none of those conditions,” Gilbert said. “Magnetic fields, radiation, chemistry, and particle-scale effects act all at once and across every scale, and the rules that work for flow in a pipe fail, sometimes badly.”
Scientists say a better understanding of turbulence could help address an array of challenges in science and engineering. Turbulence plays an essential role in how energy moves, how pollutants and heat spread through the ocean and atmosphere, how fluids move through chemical processing plants, and how plasmas behave in extreme environments.
In addition to Yale and MSU, the collaboration includes researchers from Baylor University, Auburn University, the University of Rochester, San Jose State University, and Texas A&M University-Corpus Christi. Unfunded partners include Los Alamos National Laboratory, Sandia National Laboratories, Lawrence Livermore National Laboratory, Pacific Fusion, and General Atomics.
They will combine theory, computation, AI techniques, and experimentation to build predictive models of real-world turbulence for high consequence applications.
Gilbert’s group will be the AI/machine learning glue for the center. Working with mathematicians from various institutions, the group will turn what can be proven about a physical system into constraints that a model must obey, handing back what the data suggest is true, but not yet proven. Gilbert’s group will also work with experimentalists to help decide what to measure — and how much to measure — so that data are collected in a meaningful way.
“In both directions, and across every application, the work is the same: careful analysis, benchmarking, validation, and verification of the computational methods and mathematical approximations, and their integration with the empirical results,” she said. “Only the underlying physics changes from one setting to the next.”
Manohar’s group will work with the machines these models will run on. TEMPEST’s computing needs will outgrow conventional processors, and his group will explore new computing architectures, including chips that run without a central clock and are designed together with related algorithms to run on a fraction of the power required by today’s hardware.
“The goal is predictive models of turbulence and hardware to run them that can be trusted because they are built on the physics rather than around the physics,” Gilbert said.
