Quantifying Uncertainties in Predictive Models of Inertial Confinement Fusion

Inertial confinement fusion (ICF) hydrodynamic simulations are crucial for understanding the implosion of the fuel target and are used to design future experiments at the National Ignition Facility. Typically, these simulations are computationally expensive to run. Deep learning can be used to build powerful predictive models mapping the simulation inputs (e.g. physics parameters and laser inputs) to outputs (such as neutron yield and bang time). However, most deep learning techniques yield point estimates with no information on how certain the model is in its prediction. As the model architecture increases in complexity, it becomes more unclear how to propagate and quantify uncertainties.

We present current efforts to identify the various sources of uncertainty in deep learning models trained on ICF simulation data and quantify their effects on the overall uncertainty of model-based predictions of key ICF quantities relevant for assessing the performance of the implosion. This work was performed under the auspices of the U.S. Department of Energy by Lawrence Livermore National Laboratory under contract DE- AC52-07NA27344, and released under LLNL-ABS-753983.