A major leap towards harnessing the power of the stars for clean energy on Earth has been achieved, thanks to a revolutionary artificial intelligence tool developed by scientists in the UK and Austria. The AI, named GyroSwin, can simulate the chaotic behaviour of superheated plasma inside a fusion reactor in mere seconds – a task that currently takes the world's most powerful supercomputers several days to complete.
Simulating a Star on Earth
Nuclear fusion, the process that powers the sun, promises a near-limitless source of clean energy. It works by fusing light hydrogen atoms, specifically deuterium and tritium, into helium, releasing vast amounts of energy without the long-lasting radioactive waste or greenhouse gas emissions associated with fossil fuels or nuclear fission. The challenge has always been recreating and controlling the extreme conditions needed for fusion on Earth.
This requires heating plasma – a super-hot, charged gas – to staggering temperatures of around 100,000,000°C and confining it within powerful magnetic fields inside a doughnut-shaped device called a tokamak. However, the plasma is highly turbulent, a phenomenon that causes it to leak from its magnetic cage and cool down, snuffing out the fusion reaction. The current record for sustaining a fusion reaction, held by Germany's Wendelstein 7-X device, stands at just 43 seconds.
How GyroSwin Cuts Years of Research
To create a stable, long-lasting reaction, scientists need to understand and predict plasma turbulence under countless different conditions. Traditional simulations track plasma particles in five dimensions and are incredibly computationally expensive. GyroSwin offers a transformative solution.
Developed through a collaboration between the UK Atomic Energy Authority (UKAEA), Johannes Kepler University in Linz, and the Austrian firm Emmi AI, GyroSwin is an 'AI surrogate model'. Scientists first run a small number of highly accurate, slow simulations on a supercomputer. GyroSwin is then trained on this data, learning the complex underlying physics of plasma turbulence.
Once trained, it can bypass the immense calculations and predict simulation outcomes almost instantaneously. 'GyroSwin is the first model that actually models the full plasma turbulence in all its beauty and across multiple scales,' said co-creator Dr Fabian Paischer from Johannes Kepler University.
Accelerating the Path to a Working Reactor
This dramatic speed-up is set to supercharge the engineering cycle for fusion reactors. 'Cutting turnaround from days to seconds can speed up design loops and "what if" exploration dramatically,' explained Rob Akers, Director of Computing Programmes at UKAEA. While he cautions it won't solve fusion on its own, it can 'materially speed up the engineering cycle' – a critical factor on the path to a practical power plant.
The immediate goal is to scale up GyroSwin from a proof of concept to a tool that can guide real-world experiments. It could be used to optimise existing UK facilities like the MAST Upgrade tokamak and inform the design of the UK's flagship STEP (Spherical Tokamak for Energy Production) project, which aims to build a prototype fusion power plant by the 2040s.
While a commercial fusion reactor remains a formidable challenge, breakthroughs like GyroSwin are solving fundamental problems at an unprecedented pace. By taming the turbulent heart of a miniature star with artificial intelligence, the dream of limitless, clean energy is looking more tangible than ever before.
