Manifold-informed state vector subset for reduced-order modeling

Reduced-order models (ROMs) for turbulent combustion rely on identifying a small number of parameters that can effectively describe the complexity reacting flows. With advent data-driven approaches, ROMs be trained datasets representing thermo-chemical state-space in simple systems. For low-Mach flows, full state vector serves as training dataset is typically composed temperature and chemical composition. The projected onto lower-dimensional basis evolution complex system tracked manifold. This approach allows substantial reduction transport equations to solve simulations, but quality manifold topology decisive aspect successful modeling. To mitigate challenges, several authors advocate reducing only subset major variables when ROMs. However, this often done ad hoc without giving detailed insights into effect removing certain resulting low-dimensional data projection. In work, we present quantitative manifold-informed method selecting minimizes unwanted behaviors topologies. While many past have focused species, show mixture minor species beneficial improving representations. desired effects include non-uniqueness spatial gradients dependent variable space. Finally, demonstrate improvements regressibility manifolds built from optimal opposed vector.