Species-Weighted Automated Network Reduction via Principle Component Analysis

Expedient fluid simulations of large plasma volumes with complex chemistry are hampered by the large size to which reaction networks may scale. Manual simplification of reaction networks with methods such as lumping of excited states or truncation of less-important pathways are necessary to simulate larger plasma volumes, but correspondingly reduce simulation fidelity. In this work we show a method to reduce reaction networks automatically through a combination of principal component analysis and multivariate regression which simultaneously retains information about all species. We assess the species-specific reconstruction and prediction accuracy of the method for 0D simulations of both a He/O and U/O reaction network, including a comparison of scaling and weighting methods as well as the effects of species-specific network weighting, and show the extension towards multidimensional simulations.