On the 5^th December 2023, at around 1am, an experiment in Livermore, California demonstrated energy breakeven from hydrogen fusion reactions for the first time. This experiment used the biggest and most energetic laser in the world – the national ignition facility (NIF) – to implode a millimeter-scale diamond capsule filled with a mix of deuterium and tritium and liberate 3.15MJ from fusion reactions, compared to 2.1MJ of laser energy (Q=1.5). This marks a breakthrough that was decades in the making and that has brought together teams from across the world.

The ignition shot “N231204” was notable not only because it achieved fusion “ignition”, but also because we were not surprised by the result. A week before the experiment, modeling based on past data, large-scale simulation, and statistical learning predicted that this shot design had a probability of ignition of around 50% – much greater than any previous design. This was the first time that such a prediction was made before a shot was fired and pushes the state-of-the-art for model calibration using very expensive simulation codes.

In this talk I will describe the ignition shot on NIF, and the developments that enabled it. I will dig into the details of our multiphysics simulation ensembles, deep neural network surrogates, and predictive statistical modeling, which we apply to understand the sources of performance variability in fusion experiments at NIF and to predict the outcome of future shots. Finally, I will describe the future application of these tools to the optimal design of robust ignition targets using the world’s largest exascale computing facilities.