Not so puzzling: turning a 500-wire fusion headache into a seconds long solution

ANU researcher Matthew Hole is combining maths, physics and computer science approaches to replicate the Sun’s power on earth, moving steps closer towards clean energy solutions. Instead of manually tracing 500 wires, the team used mathematical modelling to match coils to their correct connections automatically.

Fusion is the process that powers every star in the sky, and here on earth scientists have spent decades trying to recreate that energy release. Unlike nuclear fission, which splits heavy atoms apart, fusion combines light ones together, releasing enormous energy that could one day power the world with zero-carbon electricity.

Professor Matthew Hole is a plasma physics expert at the Mathematical Sciences Institute and School of Computing, in the ANU College of Systems & Society. When he arrived in southern France for a three-month visit as an International Thermonuclear Experimental Reactor (ITER) Scientist Fellow at the world’s largest fusion-energy experiment he wanted to do something practical. 

I wanted to do something that had a high impact to ITER in the timeframe that I was there. Something that mattered,” he says.

ITER, meaning “the way” in Latin, is designed to test whether we can replicate the process that powers the Sun. Inside ITER’s 30-metre-high cryostat, the plasma will be confined in a vacuum vessel and heated to more than 150 million degrees Celsius, producing enormous heat. The technology could one day drive zero-carbon electricity generation.

 “It’s literally star power on Earth,” Hole explains.

Fusion is the opposite of fission — instead of splitting heavy atoms, we combine light ones. Fusion is the same process that powers the Sun.

Turning an arduous wiring challenge into a seconds-long task

As construction of the ITER Tokamak advanced, engineers faced a formidable bookkeeping challenge.

“It was like wiring up 500 speakers in a sound system without labelling the wires and whether they are positive or negative.”  

Those sensors, about 500 in total, are critical. Without correctly identifying them, the machine simply can’t run. Engineers already had a plan to identify each coil but it was a labour-intensive task that would have taken a significant amount of time and competed with other high-priority work.    

 “That’s an extraordinary cost just to work out which coil is which,” he says. “I wanted to find a way to do it faster.” 

Where mathematics, physics and computing meet

Hole's background sits at the intersection of mathematics, physics and computing. At ITER, he combined those skills to model the magnetic-field response of every coil and then frame the puzzle as a mathematical search problem.

“I wrote out a very precise mathematical statement of what I needed to solve,” he says. “With the problem well-defined, AI assisted coding tools identified the class of problem and suggested an un-expected approach:  a Hungarian optimisation algorithm. This is a well-known technique in computer science but rarely applied to physics problems.

“If I’d tried to solve it by brute force, the calculation would grow like 2 to the power N times N-factorial – completely intractable beyond about ten coils,” Hole explains.

“The Hungarian method scales as N-cubed instead, which is manageable. My laptop could handle it.”

With a bit of coding help, he soon had a working model.

“With one week to go I demonstrated the technique worked and could identify all the coil locations and their polarities in about two seconds.”

The result was a massive reduction in time and effort.

The novel approach was very well received because it delivers the full set of coil positions and orientations in a tiny fraction of the previously allocated time. It means they can also repeat the process whenever they need to – if someone accidentally misaligns something, they can re-run it quickly.

Finding order in complexity: how AI solved a fusion puzzle

For Hole, the project was not about AI replacing human insight, but extending it.

“AI is an assistant, not a pilot,” he says. “I could have found the algorithm myself, but it would have taken weeks of searching the literature. AI accelerated that. If you write down a well-defined mathematical problem, it’s brilliant at finding existing algorithms and even helping code them.”

He describes it as a “very clever companion” for research.

“AI will help you drive the car, but it won’t tell you where to go. You’re still the pilot.”

“Mathematics is the foundation of all quantitative science,” he says simply. “You can’t do physics, chemistry or engineering without it.”

“It shows what’s possible when you combine physics insight, mathematical precision and new computational tools.”

Accelerating discovery through partnership: ANU & ITER

Dr Simon Pinches, Head of Plasma Modelling and Analysis at ITER, and a collaborator of Matthew Hole, visited Australia to strengthen links and advance collaboration established by the ITER-ANSTO Cooperation Agreement signed in 2016. He gave a public lecture at MSI about their work towards nuclear fusion and generating clean energy.

Simon Pinches has seen the same pattern play out in his own work at ITER.  In his MSI seminar, he described how AI has helped researchers untangle decades of inconsistent legacy data from earlier fusion experiments and modernise ageing scientific code.

The implications reach far beyond a single set of tangled cables. Matthew Hole’s algorithm reduces the magnet commissioning time. Each day saved brings the dream of practical fusion energy a little closer.

Ultimately, Hole says, it’s about accelerating discovery.

AI can’t replace the scientist’s curiosity or intuition,” Matthew reflects. “But it can explore scientific threads much more quickly. It’s like having tireless expert companions, across all fields, always available that can help you advance more rapidly.