Statistical inference of anomalous thermal transport with uncertainty quantification in the plasma boundary

In this study, the critical task of inferring anomalous cross-field transport coefficients is addressed in fluid-based plasma boundary simulations essential for optimal design of tokamak divertors. We developed an integrated workflow of parameter inference in the UEDGE fluid code using Bayesian Optimization (BO), which is a global optimization method suitable for optimization of computationally expensive objective functions. In the workflow, transport coefficients are inferred by minimizing a user-defined loss function, such as the discrepancy between UEDGE profile and observations. Uncertainty Quantification (UQ) is also integrated throughout the optimization to enable the efficient evaluation of the tokamak component design against plasma performance. As an example, we infer the anomalous electron thermal diffusivity $\chi_e$ from an interpretive 2-D model that includes electron conduction, convection, core temperature profile, and radiated power from bolometers, where the loss function is defined as the mean-square error of parallel heat flux at the divertor plate between UEDGE and observation from IRTV in DIII-D shots. The effectiveness of BO and UQ workflow, as well as the physical significance of interpretive 2-D model, is demonstrated.