Accelerating Chemistry and Simulations of Combustion with Neural Implicit Flow

In combustion simulations, chemistry integration and spatial resolution requirements represent significant bottlenecks. We implement variants of the neural implicit flow (NIF) method overcome limitations associated with these bottlenekcs. We use a framework, React-NIF, to integrate chemistry by mapping solutions of a subset of the thermo-chemical scalars, the representative scalars, across time increments to overcome the limitations of direct integration of complex chemistry. Another data-driven framework, based on NIF, is designed to reconstruct a superresolution of spatio-temporal data from coase solutions. The approach is designed to implement high-fidelity simulations on coarse grid where the finer resolution fields can be reconstructed from the resolved data. The two methods are illustrated using a complex fuel for chemistry acceleration and a canonical combustion problem for superresolution.