29 June 2026 to 3 July 2026
EICC, Edinburgh
Europe/London timezone

Bayesian inference of tungsten transport coefficients from laser blow-off experiments in ASDEX Upgrade

Not scheduled
20m
EICC, Edinburgh

EICC, Edinburgh

150 Morrison St, Edinburgh EH3 8EE
Poster Presentation Plasma Turbulence and Transport (MCF)

Description

With ITER’s all-tungsten plasma-facing component design, understanding impurity transport in high-temperature plasmas has become increasingly important. Impurity transport strongly affects plasma performance, as excessive impurity accumulation can degrade confinement and may even trigger disruptions.
This contribution focuses on Bayesian inference of core tungsten transport coefficients in the ASDEX Upgrade tokamak using W laser blow-off (LBO) experiments, primarily based on soft X-ray camera measurements and the Aurora transport solver [1,2]. Soft X-ray data are processed using Gaussian process tomography (GPT) [3], with hyperparameters inferred via Markov chain Monte Carlo (MCMC). The resulting tomographic reconstructions are then used for Bayesian inference of transport coefficients and boundary conditions. The forward model is provided by the 1.5D Aurora transport solver, in which the LBO is simulated starting from a steady state prior to ablation and a time-dependent source during the LBO.
This work provides a proof of concept for inferring diffusive and convective transport coefficients in the plasma core. The identifiability of these parameters is demonstrated on synthetic datasets representative of the experimental conditions. Experimental results are also presented for two W LBO discharges with different ion cyclotron resonance heating (ICRH) settings. The inferred convective transport coefficient profiles are compared with the neoclassical predictions obtained using the FACIT code [4,5,6], and the inferred diffusive coefficients are compared with the TGLF [7,8] results. The computational cost of the MCMC-based Bayesian inference is discussed, and a likelihood-free inference (LFI) [9] approach is proposed, with initial results presented.

[1] F Sciortino et al 2021 Plasma Phys. Control. Fusion 63 112001
[2] R. Dux, 2004, Habilitation Thesis, MPI-IPP
[3] Wu H et al 2024 J Fusion Energ 43 9
[4] Maget et al 2022 Plasma Phys. Control. Fusion 64 069501
[5] Fajardo et al 2022 Plasma Phys. Control. Fusion 64 055017
[6] Fajardo et al 2023 Plasma Phys. Control. Fusion 65 035021
[7] G.M Staebler et al 2025 Phys. Plasmas 12 (10)
[8] G.M. Staebler et al 2013 Nucl. Fusion 53 113017
[9] Gutmann M U and Corander J 2016 Journal of Machine Learning Research 17 1–47

Author

Jiří Malinak (Institute of Plasma Physics of the CAS, U Slovanky 2525/1a, 182 00 Prague 8, Czech Republic)

Co-authors

Clemente Angioni (Max-Planck-Institut fuer Plasmaphysik, 85478 Garching bei Muenchen, Germany) Daniel Fajardo (Max Planck Institute for Plasma Physics) Fabien Jaulmes (Institute of Plasma Physics of the Czech Academy of Sciences) Geert Verdoolaege (Ghent University) Hao Wu Jakub Seidl (Institute of Plasma Physics of the Czech Academy of Sciences) Oleg Samoylov (Max Planck Institute for Plasma Physics) Roberto Bilato (Max Planck Institute for Plasma Physics, Germany)

Presentation materials