An efficient methodology to analyze plasma edge model parameter sensitivities

Abstract

Quantifying uncertainties on code outputs is an important step for code-based design and scenario development. Because of the high computational cost of plasma edge transport simulations, the propagation of uncertainties on input parameters quickly becomes intractable. The paper starts with a short overview of current concepts to deal with this issue. A practical in-parts adjoint approach to sensitivity calculation is then applied for computationally efficient uncertainty propagation. The cumbersome derivation-by-hand of the sensitivity expressions is avoided, while the computational cost is roughly kept independent of the number of uncertain parameters. Exemplarily sensitivities of the outer strike point temperature and a heat load objective are calculated for a WEST case. Transport coefficients, boundary condition parameters, rate coefficients, as well as uncertain parameters in the magnetic equilibrium calculation are considered. The sensitivities are verified to be accurate, while the computational cost to compute the entire sensitivity matrix is equivalent to only one additional plasma edge simulation for each output quantity of interest. Furthermore, several logical trends are observed in the sensitivities.