A Symbolic Computational Abstraction of Chemistry Libraries

Accurate simulations of reacting flows require computational representations of chemical source terms. Dedicated libraries can provide such representations by acting as interfaces between mechanistic data and flow solvers. However, libraries form a fragmented computational landscape, with each library typically implementing kinetic models for different applications (e.g., combustion, plasma, or nuclear reactions). Furthermore, libraries impose rigid data layout constraints and must be compiled before linking with the flow solver. These issues complicate the simulations of multi-kinetic flows (e.g., plasma-coupled combustion) and hinder the automatic differentiation (AD) and joint optimization of flow solvers on modern hardware, such as GPUs. We develop a symbolic representation of chemistry that addresses these issues. This representation uses the general structure of mass-action kinetics, making the system agnostic to the underlying library. Chemical and computational details are progressively added using domain-specific library data to create an application-specific computational model. The computational model is well-suited for generating code for accelerator devices. We demonstrate this property by using three different libraries to compute combustion, plasma, and nuclear kinetics on GPU devices.