GyroSwin: 5D Surrogates for Gyrokinetic Plasma Turbulence Simulations
Fabian Paischer, Gianluca Galletti, William Hornsby, Paul Setinek, Lorenzo Zanisi, Naomi Carey, Stanislas Pamela, Johannes Brandstetter
2025-10-10
Summary
This paper introduces a new computer model, called GyroSwin, designed to better understand and predict the behavior of plasma in nuclear fusion reactors, ultimately aiming to make fusion energy a practical reality.
What's the problem?
Nuclear fusion requires extremely hot plasma, but this plasma is unstable and prone to turbulence. This turbulence causes energy to leak out, making it hard to sustain the reaction. Scientists use complex equations to model this turbulence, but these equations are incredibly difficult and time-consuming to solve with computers. Existing simplified models, while faster, miss important details about how the turbulence actually works, leading to inaccurate predictions.
What's the solution?
The researchers created GyroSwin, a new type of computer model based on artificial intelligence. It's designed to handle the full complexity of the equations governing plasma turbulence in five dimensions, something previous models couldn't do efficiently. GyroSwin uses a technique inspired by image recognition, adapting it to work with this complex plasma data, and it specifically focuses on capturing the interactions between different parts of the plasma. It's also designed to be scalable, meaning it can become more accurate as computing power increases.
Why it matters?
GyroSwin is a significant step forward because it's much faster than traditional methods for simulating plasma turbulence – reducing the computational cost by a factor of a thousand – while still being accurate and capturing the important physics. This allows scientists to design better fusion reactors that can contain the plasma more effectively, bringing us closer to a clean and sustainable energy source.
Abstract
Nuclear fusion plays a pivotal role in the quest for reliable and sustainable energy production. A major roadblock to viable fusion power is understanding plasma turbulence, which significantly impairs plasma confinement, and is vital for next-generation reactor design. Plasma turbulence is governed by the nonlinear gyrokinetic equation, which evolves a 5D distribution function over time. Due to its high computational cost, reduced-order models are often employed in practice to approximate turbulent transport of energy. However, they omit nonlinear effects unique to the full 5D dynamics. To tackle this, we introduce GyroSwin, the first scalable 5D neural surrogate that can model 5D nonlinear gyrokinetic simulations, thereby capturing the physical phenomena neglected by reduced models, while providing accurate estimates of turbulent heat transport.GyroSwin (i) extends hierarchical Vision Transformers to 5D, (ii) introduces cross-attention and integration modules for latent 3Dleftrightarrow5D interactions between electrostatic potential fields and the distribution function, and (iii) performs channelwise mode separation inspired by nonlinear physics. We demonstrate that GyroSwin outperforms widely used reduced numerics on heat flux prediction, captures the turbulent energy cascade, and reduces the cost of fully resolved nonlinear gyrokinetics by three orders of magnitude while remaining physically verifiable. GyroSwin shows promising scaling laws, tested up to one billion parameters, paving the way for scalable neural surrogates for gyrokinetic simulations of plasma turbulence.