Using neural posterior estimators to calibrate and propagate constitutive data uncertainties for resistive magnetohydrodynamic liner simulations
Physics of Plasmas · 01.11.2026 · 33 (11)
DOI: 10.1063/5.0347805 ↗Kurz gefasst
Fachartikel in der Zeitschrift Physics of Plasmas, online erschienen am 12.11.2026. Verfasst von J. R. Allison und 11 weiteren Autor:innen. Die Arbeit nennt 99 Literaturangaben. Der Volltext ist frei zugänglich (Open Access, CC BY 4.0).
Abstract (Originalsprache)
Electrical conductivity and equation-of-state (EoS) models are key inputs to magnetohydrodynamic (MHD) simulations of imploding liners, but uncertainties in these constitutive data can strongly influence predicted current transport, heating, compression, and melt dynamics. We present a Bayesian simulation-based inference framework, based on neural posterior estimators, for calibrating and propagating uncertainties in constitutive data used in hydrodynamic simulations of liner implosions. The method proceeds in two stages. First, parametric uncertainties in conductivity and EoS models are calibrated against literature data assembled in First Light Fusion's Physics Assurance Database (PAD) to define physically informed priors. Second, these uncertainties are propagated through ensembles of coupled circuit-MHD simulations and conditioned against fiber-optic current-sensor and photonic Doppler velocimetry measurements from a pulsed-power-driven aluminum liner experiment. Neural posterior estimators are used to perform likelihood-free inference over both constitutive and load-coupling parameters. For the experimental case considered, including calibrated constitutive uncertainty reduces the inferred liner-current fraction by approximately 5% and delays the median melt-wave burnthrough time by approximately 90 ns, while broadening both posterior distributions. The aluminum constitutive parameters themselves are only weakly updated beyond the PAD-informed prior. These results demonstrate the importance of propagating independently calibrated constitutive uncertainty through integrated MHD inference, where it can materially influence inferred load coupling and melt dynamics even when the constitutive parameters themselves are only weakly updated by the experimental data.
Zitieren
APA
Allison, J. R., Chapman, D. A., Edwards, T., Fraser, A. R., Joiner, N., Kosteletos, T., Mohamed, J. I., Pecover, J. D., Turner-Wilson, I., Nash, O., Burdiak, G. C., & Bordas, R. (2026). Using neural posterior estimators to calibrate and propagate constitutive data uncertainties for resistive magnetohydrodynamic liner simulations. Physics of Plasmas, 33(11). https://doi.org/10.1063/5.0347805
Vancouver
Allison JR, Chapman DA, Edwards T, Fraser AR, Joiner N, Kosteletos T, et al. Using neural posterior estimators to calibrate and propagate constitutive data uncertainties for resistive magnetohydrodynamic liner simulations. Physics of Plasmas. 2026;33(11). doi:10.1063/5.0347805
BibTeX
@article{Allison2026,
author = {Allison, J. R. and Chapman, D. A. and Edwards, T. and Fraser, A. R. and Joiner, N. and Kosteletos, T. and Mohamed, J. I. and Pecover, J. D. and Turner-Wilson, I. and Nash, O. and Burdiak, G. C. and Bordas, R.},
title = {Using neural posterior estimators to calibrate and propagate constitutive data uncertainties for resistive magnetohydrodynamic liner simulations},
journal = {Physics of Plasmas},
year = {2026},
volume = {33},
number = {11},
doi = {10.1063/5.0347805}
}
EUVIMED