latentBatchSMOKE++, a latent variable transport solver for batch reactors with detailed chemistry computation

Combustion is a complex, multi-scale process involving intricate physical phenomena that make its simulation highly demanding in terms of computational resources. To address these challenges, reduced-order models are critical for both enhancing our understanding and accelerating simulations. This study presents a latent variable transport framework tailored for efficient simulation of 0D batch reactors. To further improve solver performance, a projected tolerance strategy is introduced, offering robustness across varying score scalings. The approach is tested on the combustion of methane, propane, and n-heptane, using both detailed and lumped kinetic mechanisms. Results show that projected tolerances outperform fixed tolerances by preserving accuracy while significantly reducing computation time and the number of solver steps. The influence of the kinetic mechanism type is also examined: the latent solver yields greater efficiency gains with detailed mechanisms, owing to their higher species count and lower stiffness, whereas lumped mechanisms, despite being simpler, present higher stiffness that limits computational benefits. Overall, the proposed method provides an effective balance between accuracy and efficiency, and represents a valuable strategy for reduced-order modeling in combustion research.