Bayesian Inference for Plasma Temperature and Density from Emission Spectroscopy

In emission spectroscopy, it is common to infer plasma temperature and

density by fitting a model spectrum to observations. In this talk, we

develop a Bayesian inference approach to this problem. The primary

innovation of this work is a technique that accounts for the possible

discrepancy between the mathematical model and the observations.

Specifically, we pose a typical algebraic model of the spectrum as a

sum of lines, each with an associated position, strength, and width.

The discrepancy model is based on plausible perturbations of this

algebraic model---i.e., perturbations of the position, strength, and

width parameters. These plausible perturbations are characterized by

a statistical model, the parameters of which are inferred

simultaneously with the plasma temperature and density, so that they

account for actual discrepancies between the observed and modeled

spectrum. By constructing the discrepancy model in this way it is

able to account for mismatches between the modeled and observed

spectra that may arise due to experimental noise, contaminants in the

plasma, and modeling errors. This protects against overly certain

inferences for the parameters of interest, leading to more realistic

uncertainty in the inferred quantities. As an example, the process is

applied to a series of observed spectra recorded at the Plasma Liner

Experiment (PLX) facility.