High-energy density hohlraum design using forward and inverse deep neural networks (original) (raw)

Section snippets

Experimental design for opacity measurements

Hohlraums are used in High Energy Density Physics (HEDP) and Inertial Confinement Fusion (ICF) experiments to convert laser energy to thermal x-rays for imploding capsules, heating targets, and generating thermal radiation waves [1]. The physics of lasers, wall heating, and ablation are complex and often not well modeled. Also, the complete set of physical phenomena that are understood to be present in an experiment are not always included in the simulation codes. Even with these

Simulations of laser-driven hohlraums

Our simulation strategy followed the procedure detailed in [3]. The simulations utilized two laser sources, representing NIF beam cones oriented at 44.5∘ and 50∘ from the hohlraum axis. Both source descriptions incorporated NIF phase-plate data in order to reproduce as closely as possible an empirical beam intensity pattern in the plane of best focus (i.e., transverse to the beam propagation direction). The central elliptical high-intensity regions had semi-axes 635×367 μm and 593×343 μm, while

Forward and inverse deep neural networks

From each simulation in the ensemble the output we receive is the Dante measurement of the radiation temperature as a function of time, TD(t;θ), where θ=(scale,sc_length,Rapt,pulse_length). The times output by the simulation are not at equally-spaced intervals; therefore, we use cubic spline interpolation to put them at a unified set of 50 time points. For training the model we also scale the time variable by the inverse of the laser pulse scale factor so that the time variable gives a position

Forward model

From the forward model we can investigate how changing the hohlraum shape parameters and laser pulse affects the radiation temperature as a function of time. In Fig. 5 we show the effects of scale, sample chamber length, and laser pulse length. The effect of aperture radius, Rapt, is not shown in the figure because the radiation temperature sampled by the Dante-1 field of view is effectively independent of this parameter. This is to be expected, as the aperture controls radiation exposure to

Application of the statistical Hohlraum model to opacity measurements

The ultimate goal for this statistical model, which we have demonstrated here on a simple hohlraum, is to guide and accelerate the design of the more sophisticated hohlraum used in opacity measurement experiments on the National Ignition Facility [31], [32]. As the Opacity-on-NIF program seeks higher sample temperatures and densities, new hohlraum designs will require additional parameter optimization, such as reducing the hohlraum emission that detrimentally increases spectrometer backgrounds,

Conclusion

We have demonstrated the training and use of forward and inverse deep neural network models to predict hohlraum behavior and to design hohlraums for HEDP and ICF experiments. We also analyzed iron opacity data for the modeled conditions and motivated a future inclusion of theoretical opacity data as priors in the models.

With an inexpensive neural network model for the Dante response, a panoply of tools become available to an experiment designer that would be prohibitively costly using

CRediT authorship contribution statement

Ryan McClarren: Conceptualization, Methodology, Writing - original draft, Writing - review & editing, Supervision, Drafting credit statements. I.L. Tregillis: Conceptualization, Methodology, Investigation, Writing - review & editing. Todd J. Urbatsch: Conceptualization, Writing - original draft, Writing - review & editing, Project administration. E.S. Dodd: Conceptualization, Writing - review & editing.

Declaration of Competing Interest

The authors declare that they have no known competing financial interests or personal relationships that could have appeared to influence the work reported in this paper.

Acknowledgement

The authors would like to thank the anonymous reviewer for the insightful comments that led to a much improved final version. This work was supported by the US Department of Energy through the Los Alamos National Laboratory (LANL). LANL is operated by Triad National Security, LLC, for the National Nuclear Security Administration of the US DOE (Contract No. 89233218CNA000001), LA-UR-20-26126.