An efficient framework for reliable calibration and uncertainty estimation of novel self-consistent cross-field transport models for plasma edge simulations of fusion devices

Plasma edge codes are the work-horse for interpretation of experiments and design of fusion devices. One of the largest uncertainties in these codes is due to the approximation of the cross-field turbulent transport with ad-hoc diffusion models, hampering the codes’ predictive capabilities. Recently, self-consistent cross-field transport models were proposed to reduce this uncertainty. However, rigorous calibration and comparison of the competing transport models accounting for both modelling and experimental errors is currently missing. 
The first aim of this project is to develop a reliable methodology for plasma edge model calibration. I will use a Bayesian approach to consistently integrate all data from heterogenous diagnostics into the calibration problem, dealing with error cross-correlations, partially redundant information, uncertainty in the magnetic equilibrium, and data outliers. 
Computationally expensive plasma edge codes require efficient strategies to solve the calibration problem. I will develop an efficient one-shot method to solve the calibration problem through adjoint-based optimization, exploiting Algorithmic Differentiation to automatically compute sensitivities. With the Laplace method, I will provide evidence estimates to compare the novel transport model against the diffusive model, and provide uncertainty estimates on the calibrated parameters.
The project will set a significant step towards validated plasma edge models for reliable reactor design.