Bayesian equilibria of axisymmetric plasmas.

Bayesian models of axisymmetric plasmas using Gaussian processes and force balance equations have been developed. These models give the full joint posterior probability distributions over plasma current distributions and pressure profiles given the magnetic field and pressure measurements simultaneously. The toroidal currents such as plasma and magnetic field coil currents are modelled as a grid of toroidal current carrying solid beams. The plasma pressure and poloidal current flux profiles are given as a function of the poloidal magnetic flux surface, determined by the toroidal currents. Inference of all these physics parameters is a tomographic problem, thus, in order to exclude unreasonable solutions, two different prior distributions have been exploited: a Gaussian process prior and an equilibrium prior. The Gaussian process prior constrains the plasma current distributions by their covariance (smoothness) function whose hyperparameters have been optimally selected by Bayesian Occam's razor. On the other hand, the equilibrium prior imposes the magnetohydrodynamic force balance by introducing observations that the differences between the magnetic force and the plasma pressure gradient are almost zero at every plasma current beam. These \textit{virtual} observations emphasise equilibrium solutions \textit{a~priori} as a part of the prior knowledge. These models with the two different priors employ predictive models of magnetic sensors and other plasma diagnostics in order to find all possible solutions consistent with all the measurements. The complex, high dimensional posterior distributions are explored by a new method based on the Gibbs sampling scheme.