Development of a Bayesian method for the analysis of inertial confinement fusion experiments on the NIF

Abstract

The complex nature of inertial confinement fusion (ICF) experiments results in a very large number of experimental parameters that are only known with limited reliability. These parameters, combined with the myriad physical models that govern target evolution, make the reliable extraction of physics from experimental campaigns very difficult. We develop an inference method that allows all important experimental parameters, and previous knowledge, to be taken into account when investigating underlying microphysics models. The result is framed as a modified χ 2 analysis which is easy to implement in existing analyses, and quite portable. We present a first application to a recent convergent ablator experiment performed at the NIF, and investigate the effect of variations in all physical dimensions of the target (very difficult to do using other methods). We show that for well characterised targets in which dimensions vary at the 0.5% level there is little effect, but 3% variations change the results of inferences dramatically. Our Bayesian method allows particular inference results to be associated with prior errors in microphysics models; in our example, tuning the carbon opacity to match experimental data (i.e., ignoring prior knowledge) is equivalent to an assumed prior error of 400% in the tabop opacity tables. This large error is unreasonable, underlining the importance of including prior knowledge in the analysis of these experiments. (J.A. Gaffney) cluded, along with prior work on microphysics models, in 20 a consistent and efficient analysis. The approach has been 21 designed to couple with existing radiation-hydrodynamics 22 42 of important variables and prior knowledge has a dra-43 matic effect on the results of inference (even if they are 44 purely sources of noise). Our method allows these effects 45 to be included with almost no computational overhead. 46 We demonstrate this by presenting an analysis of a single 47 NIC convergent ablator (conA) shot [8, 9], N110625. We 48 find that variations in the dimensions of the target can 49 have a dramatic effect on the inferred drive and carbon 50 opacity, although this is mitigated by thorough metrology 51 of the target. Our method also allows prior knowledge to 52 be included and in the case of ICF we find that this is an 53 extremely important factor. We use it to investigate the 54 implied error in microphysics models associated with ne-55 glecting this prior work. The inclusion of this prior knowl-56 edge is an important strength of the Bayesian method, as 57 it provides context for observed data and therefore allows 58 meaningful information to be inferred even from a sin-59 gle experimental result. The total information from a set 60 of experiments can be viewed as a series of such single-61 shot inferences, allowing the analysis performed here to 62 be generalised to full experimental campaigns very easily. 63 The approach is to treat the output of the simula-64 tion code as probabilistic, and to apply standard meth-65 ods of Bayesian analysis ...

The data-driven future of high-energy-density physics

“…AI tools have been applied to the automatic analysis and featurization of complex data types like spectra, images 54,105 , and line-of-sight dependent quantities. There has been significant interest in using Bayesian inference to improve diagnostics 106 , and to synthesize observations in both focused HEDP experiments 51 and full-scale ICF experiments 50,[107][108][109] . The ultimate aims of using these methods to improve physics understanding, and the reliability of simulations in extrapolating to new designs 48 or facilities, have been addressed though machine learning 42,110 , Bayesian model calibration 50 , and transfer learning 58,111 .…”

Quantifying equation-of-state and opacity errors using integrated supersonic diffusive radiation flow experiments on the National Ignition Facility

“…29 and 30. Such involved and rigorous statistical assessments are rarely made in high energy-density (HED) experiments (Gaffney et al 31,32 being recent notable exceptions); this work quantifies uncertainties to a higher degree than is standard in the field of radiation wave propagation, such as Afshar-Rad et al, 9 Back et al, 3,4 Bozier et al, 10 Gregori et al, 7 Hoarty et al, 11,12 Massen et al, 8 and Willi et al 13 Furthermore, this work assesses the probability that data can be described sufficiently by the named models.…”

$$\mathrm{ND}^2\mathrm{AV}$$ ND 2 AV : N-dimensional data analysis and visualization analysis for the National Ignition Campaign

“…This modification has been developed in order to blur the line between macro-and micro-variables since the experimental observables (usually the same in both cases) clearly depend on both types simultaneously. A description of the Bayesian approach, and the simulations themselves, can be found in [13] and [12].…”