Chemical complexity in astrophysical simulations: optimization and reduction techniques

TL;DR

This paper introduces optimization and reduction techniques for chemical networks in astrophysical simulations, significantly improving computational efficiency while preserving essential chemical information.

Contribution

It presents novel buffer, flux-based, and topological methods for reducing chemical network complexity and computational cost in astrophysical simulations.

Findings

Achieved 2 to 5 times speed-up in simulations.

Developed hybrid methods combining buffer and flux strategies.

Provided a topological approach for network reduction.

Abstract

Chemistry has a key role in the evolution of the interstellar medium (ISM), so it is highly desirable to follow its evolution in numerical simulations. However, it may easily dominate the computational cost when applied to large systems. In this paper we discuss two approaches to reduce these costs: (i) based on computational strategies, and (ii) based on the properties and on the topology of the chemical network. The first methods are more robust, while the second are meant to be giving important information on the structure of large, complex networks. To this aim we first discuss the numerical solvers for integrating the system of ordinary differential equations (ODE) associated with the chemical network. We then propose a buffer method that decreases the computational time spent in solving the ODE system. We further discuss a flux-based method that allows one to determine and then cut on the fly the less active reactions. In addition we also present a topological approach for selecting the most probable species that will be active during the chemical evolution, thus gaining information on the chemical network that otherwise would be difficult to retrieve. This topological technique can also be used as an a priori reduction method for any size network. We implemented these methods into a 1D Lagrangian hydrodynamical code to test their effects: both classes lead to large computational speed-ups, ranging from x2 to x5. We have also tested some hybrid approaches finding that coupling the flux method with a buffer strategy gives the best trade-off between robustness and speed-up of calculations.