Perspective on Fusion Energy with Lasers

Can Machine Learning help ease the arduous path toward a viable solution?

Nuclear fusion holds huge promise as a source of clean, abundant energy that could power the world. Now, researchers at a national laboratory in the US have achieved something physicists have been working towards for decades. But just how close are we to producing commercial energy from fusion? This is an aspiration that has stimulated the minds of great people, but the path towards an economically viable fusion power plant is an arduous one. Are we at a better point in time to ensure that dream is finally achieved?

The idea of laser-driven fusion was first put forward by John Nuckolls in 1972, close to half a century ago, and, just very recently, a major milestone has been achieved at the world's largest laser facility “The National Ignition Facility (NIF)” at the Lawrence Livermore National Laboratory (LLNL). There, for the first time, scientists have demonstrated that the amount of energy released by fusion products exceeds the energy used to initiate the thermonuclear reactions, which we call break-even. NIF's approach is to mimic as closely as possible what happens in the Sun, nature’s very effective nuclear fusion reactor. The fuel, made of isotopes of hydrogen, is compressed to enormous densities such that fusion reactions can occur more easily. However, as we experience in everyday life, just like we are trying to squeeze a squishy ball, it can squirt out in all directions. The Sun prevents this from occurring by using its own gravity. At NIF, the trick is to compress fast enough that the chain reaction takes places before the high density cores disassemble.

NIF's approach is not the only one, and many others have been pursued, both by government and private organizations. Some promise better efficiency and energy gain - which are a necessary requirement for a commercial reactor on the grid sometime in the future. A comparison that has been often made is that nowadays fusion reactors are not that different from the Wright Brothers' first airplane. It did fly, but it is also disparate from any present-day commercial aircraft. So is NIF from what may be a future fusion power plant. The question is: how can we get there faster? The plot in the figure above is very telling. It shows the fusion yield (a measure of the energy produced vs the laser energy used to compress) since NIF was commissioned. While different colours correspond to different designs, one fact is immediately noticeable: in just the last few years there has been enormous progress.

How has this been possible? The answer is easy: machine learning. Success of fusion energy relies on tuning many different knobs, but the problem is that changing one affects all the others, so this makes tuning extremely hard. To add to this, numerical simulations are hardly predictive. For example, it is not widely known that the August 8, 2021 impressive result of 1.3 MJ of neutron yield came as a surprise that was not predicted by codes. Machine learning is good at extracting latent correlations even with large uncertainties. Not only this, machine learning can also extrapolate to new scenarios and make informed predictions. This is why we have seen such successes in recent years. NIF's case is not unique and many fusion industries are now following similar approaches.

Finally, with ignition now achieved on NIF, not only fusion energy is unlocked, but also a door is opening to new science. Inertial fusion-related research promises a four-order-of-magnitude increase in thermal neutron source brightness, thereby revolutionising neutron scattering for applications across the natural sciences, from biochemistry to life sciences. Fundamental physics of supernovae explosions and radiation dominated plasmas is now accessible and directly testable in the laboratory.

As many are aware, the threats facing humankind from climate change are urgent, and specific milestones have been outlined by the UN for countries to achieve within the next 10 to 30 years. It may well be that inertial fusion energy is not yet well-developed enough to ever be in the first wave of technologies to have an impact on those relative short terms, but however the future pans out, the overall goal of a clean, safe, and abundant source of nuclear energy will remain. With the ever increasing reliance on machine learning there is the concrete hope that fusion energy may become a reality even earlier than that.

Main figure credit: Mark Herrmann/LLNL/Physics Today (DOI:10.1063/PT.6.2.20221213a)
Gianluca Gregori, Co-founder
Gianluca advises our research and machine learning algorithm development team, focusing on the technological and business applications for the clean energy market. He’s also a senior faculty member in the Department of Physics at Oxford University and runs a research group in fusion energy and laboratory astrophysics.

Latest Articles

From Unknown Unknowns to Known Unknowns

Unravelling the beauty of uncertainty

Perspective on Fusion Energy with Lasers

Can Machine Learning help ease the arduous path toward a viable solution?

Days to Seconds

A new lens – imagine the future possibilities