A Kernel-Embedding Framework for Inferring Generic Electron Energy Distributions from Spectroscopic Diagnostics
Abstract
Inferring more exotic electron energy distribution functions (EEDFs) from emission or absorption spectra faces potential physical limitations, and complexity challenges, but is crucial for meeting certain AFRL objectives. A forward model plus Markov Chain Monte Carlo (MCMC) over parameters influencing the EEDF forms a common strategy for confronting the problem. This strategy automatically incorporates physical constraints, but involves situation-specific knowledge (or assumptions). While we anticipate utilizing problem-specific knowledge, and any other available diagnostics in ultimate use cases, approaching the problem from the agnostic limit proved fruitful. Besides letting us examine model and diagnostic assumptions, it spurred developing a more flexible EEDF representation, and framework for inference. Specifically, we generalized EEDF shape through splines or weighted basis functions in cumulative-distribution function (CDF) space, and performed inference via data-driven kernel-embedding techniques - as well as a more conventional method combining annealing with MCMC. The kernel-embedding approach accommodates problem specific and agnostic EEDFs, facilitates faster inference, and can complement more conventional MCMC. Crucially, it also provides a handle on the messiness of the problem.
Document Details
- Document Type
- Technical Report
- Publication Date
- Mar 18, 2022
- Accession Number
- AD1173560
Entities
People
- David L. Bilyeu
- Hal J. Cambier
Organizations
- Air Force Research Laboratory