There are plenty of reasons why fusion energy has yet to become reality, but according to a group of researchers from the University of Texas at Austin and their collaborators, we may be one modeling breakthrough closer.
The team’s paper introduces a new method to address a long-standing problem in fusion physics: high-energy particles, especially alpha particles produced by fusion reactions, don’t always stay confined within the magnetic fields designed to hold them. In stellarators, the type of fusion reactor examined in the study, poor confinement of these fast particles can lead to significant energy loss, making sustained fusion much harder to achieve.
Stellarators use external magnetic coils to generate complex magnetic fields that confine plasma inside the reactor. But those fields can struggle to contain fast-moving particles, especially fusion-born alpha particles, whose trajectories don’t always conform to the assumptions built into standard field models.
Accurately predicting how these particles move often requires full-orbit simulations based on the Lorentz force – essentially tracking each particle’s exact path. That level of modeling is extremely precise, but also computationally expensive.
UT Austin noted that it could take thousands of designs with slight adjustments to eliminate all the holes in a stellarator magnetic field, making precision design of fusion reactor magnetic fields all but impossible.
Instead of simulating every particle’s full orbit, scientists traditionally rely on perturbation theory, a method that approximates particle motion by building on simpler, ideal cases. While effective in many situations, it breaks down for fast-moving particles in complex magnetic fields, leading to inaccurate confinement predictions and underperforming reactor designs.
“Direct application of Newton’s laws is too expensive. Perturbation methods commit gross errors,” said lead paper author Josh Burby, who is an Assistant Professor in the UT Austin physics department. “Ours is the first theory that circumvents these pitfalls.”
While deeply technical, the work boils down to improving on standard approximations with a machine-learned model that better tracks fast-moving particles in the complex magnetic fields of stellarators, by leveraging a hidden symmetry that traditional methods struggle to capture outside ideal conditions.
According to a UT Austin spokesperson, the new method is ten times faster than Newtonian analysis without losing any of the accuracy, but still has shortcomings. The paper stated that the new data-driven method “falls short” in certain applications because it has to be repeated whenever the magnetic field changes, “encurring [sic] unacceptable computational overhead.”
As with all things lately, we may have to rely on AI to be the saving grace that could make the method practical.
“Recent advances in machine learning, especially in foundation models and sparse regression, stand poised to tackle this challenge,” the researchers said in the conclusion of their paper.
Additionally, the “non-perturbative guiding center theory,” as it’s called in the paper, is developed with stellarators in mind, but the approach could, in principle, be extended to tokamaks – the more widely used fusion reactor design.
Unfortunately, the larger problem with fusion energy remains that, in practice, it still takes more energy to run the whole system than the reaction gives back. One approach, firing high-powered lasers at a tiny fuel pellet, has shown real promise.
In the one case where researchers saw so-called “ignition,” meaning the energy released by the fusion reaction exceeded the energy delivered to the fuel itself. But even then, far more energy was needed to power the lasers and supporting systems than the fusion produced, resulting in a net energy loss when looking at the full input-output balance.
We definitely need more energy to power our increasingly electric-hungry society, including the massive datacenters that train AI systems. But as with every advance of this type, it’s just one drop in a gargantuan bucket of other things that have yet to be resolved. ®
