Scientists Just Took a Giant Step Toward Scaling Up Nuclear Fusion: ‘What We’ve Done Here Is the Start of What Is Still a Long Journey’

A team of researchers at MIT think they may have lowered one of the major barriers to achieving large-scale nuclear fusion—taking us one step closer to making an abundant form of energy a reality.

By harnessing the same processes that power stars, we would have access to a clean, safe, and practically limitless energy source. Scientists have built reactors to try and tame fusion, with one of the most explored being the tokamak. Essentially a donut-shaped tube that uses strong magnets to confine the plasma needed to power fusion reactions, the tokamak has shown great potential. But to fully realize that, scientists must first navigate the potential pitfalls that such energy carries with it, including how to slow down a fusion reaction once it is in progress.

That’s where the new research^[1] comes in: Using a combination of physics and machine learning, the researchers predicted how the plasma inside a tokamak reactor would behave given a set of initial conditions—something that researchers have long puzzled over (it is hard to look inside a fusion reactor mid-run, after all). The paper was published Monday in Nature Communications.

“For fusion to be a useful energy source, it’s going to have to be reliable,” Allen Wang, study lead author and a graduate student at MIT, told MIT News^[2]. “To be reliable, we need to get good at managing our plasmas.”

With great power comes great risks

When a tokamak reactor is fully running, the plasma current inside can circulate at speeds of up to about 62 miles (100 kilometers) per second and at temperatures of 180 million degrees Fahrenheit (100 million degrees Celsius). That is hotter than the Sun’s core.

If the reactor has to be shut down for any reason, operators initiate a process to “ramp down” the plasma current, slowly de-energizing it. But this process is tricky, and the plasma can cause “scrapes and scarring to the tokamak’s interior—minor damage that still requires considerable time and resources to repair,” the researchers explained.

“Uncontrolled plasma terminations, even during rampdown, can generate intense heat fluxes damaging the internal walls,” explained Wang. “Quite often, especially with the high-performance plasmas, rampdowns actually can push the plasma closer to some instability limits. So, it’s a delicate balance.

Indeed, any misstep in operating fusion reactors can be costly. In an ideal world, researchers would be able to run tests in working tokamaks, but because fusion is still not efficient, running one of these reactors is incredibly costly, and most facilities will only run them a few times a year.

Looking to the wisdom of physics

For their model, the team found a delightfully clever method^[3] to overcome the limitations in data collection—they simply went back to the fundamental rules of physics. They paired their model’s neural network with another model describing plasma dynamics, and then trained the model on data from the TCV, a small experimental fusion device in Switzerland. The dataset included information about variations in the plasma’s starting temperature and energy levels, as well as during, and at the end of each experimental run.

From there, the team used an algorithm to generate “trajectories” that laid out for the reactor operators how the plasma would likely behave as the reaction progressed. When they applied the algorithm to actual TCV runs, they found that following the model’s “trajectory” instructions were perfectly able to guide operators to ramp the device safely down.

“We did it a number of times,” Wang said. “And we did things much better across the board. So, we had statistical confidence that we made things better.”

“We’re trying to tackle the science questions to make fusion routinely useful,” he added. “What we’ve done here is the start of what is still a long journey. But I think we’ve made some nice progress.”