In the quest to harness the power of fusion energy, scientists are constantly grappling with complex phenomena that can disrupt the process. One such challenge is the runaway electron (RE) avalanche, a rapid and uncontrolled growth of high-energy electrons that can damage tokamak reactors, the most advanced devices for producing fusion energy. A recent study published in the journal *Nuclear Fusion* (translated from Latin as “Nuclear Fusion”), led by Jonathan S. Arnaud from the University of Florida, introduces a groundbreaking approach to predict and potentially mitigate this issue using physics-informed machine learning.
Arnaud and his team have developed a physics-constrained deep learning surrogate that can predict the exponential growth rate of runaway electrons in a plasma containing partially ionized impurities. This surrogate is based on a physics-informed neural network (PINN) that learns the adjoint of the relativistic Fokker–Planck equation in steady-state. “This approach enables a rapid prediction of the RE avalanche for a broad range of plasma parameters, which is crucial for understanding and managing tokamak disruptions,” Arnaud explains.
The significance of this research lies in its potential to accelerate the development of an integrated description of tokamak disruptions. By embedding a steady-state power balance equation and atomic physics data directly into the PINN, the model is constrained to physically consistent temperatures and charge state distributions. This not only enhances the accuracy of predictions but also drastically reduces the computational cost of training the model.
One of the novel aspects of this study is the development of a new closure for the relativistic electron population. This closure improves the accuracy of evaluating the secondary source of runaway electrons compared to the traditional Rosenbluth–Putvinski source. “Our novel closure enables improved accuracy, which is essential for the reliable prediction of RE avalanche growth rates,” Arnaud notes.
The avalanche surrogate was verified against Monte Carlo simulations, demonstrating its ability to accurately predict the RE avalanche growth rate across a wide range of plasma parameters. This includes scenarios encompassing distinct tokamak disruption events, highlighting the model’s versatility and robustness.
The implications of this research for the energy sector are profound. By providing a rapid and accurate prediction of runaway electron avalanches, this model can help in the design and operation of safer and more efficient tokamak reactors. This, in turn, can accelerate the commercialization of fusion energy, a clean and virtually limitless power source.
Arnaud’s work represents a significant step forward in the field of fusion energy research. By combining physics-based models with advanced machine learning techniques, his team has opened up new avenues for understanding and managing tokamak disruptions. As the world continues to seek sustainable and reliable energy solutions, this research offers a promising path towards a future powered by fusion energy.