Regime-Aware Model Selection for Reactive Transport with Asymptotic Completion

Andrew R. Garcia · ChemRxiv (2026) · DOI: 10.26434/chemrxiv.15001766/v1

Resumen

Selecting an inappropriate transport model can introduce large errors in reactive flow prediction. We propose a regime-aware model selection framework for convection–diffusion–reaction (CDR) systems governed by the Péclet (Pe) and Damköhler (Da) numbers. The framework automatically identifies which combination of advection, diffusion, and reaction operators is physically valid for a given regime, and constructs stable reference trajectories using classical perturbation theory when standard numerical integration fails. At small Pe, a matrix-exponential solver provides an unconditionally stable reference via regular perturbation. At large Pe, a Van Dyke matched asymptotic expansion yields a composite solution with a boundary layer of thickness O(1/Pe). A stiff Radau solver handles high-Da nonlinearity where the asymptotic approximation breaks down. Model selection is performed by minimizing a joint residual-plus-rollout score against the regimeappropriate reference trajectory. Experiments on a 4 × 4 Pe-Da parameter grid and a cylindrical reactive-wall benchmark produce clean, physically interpretable regime maps with finite confidence margins across all conditions. A chemical source detection scenario shows that selecting the wrong model introduces source estimation errors exceeding four orders of magnitude, while the proposed framework stays below 0.25% error across all tested regimes, compared to a naive diffusion-only baseline that reaches 16,532%.

Instituciones: Universidad de Ingeniería y Tecnología