Emulating the interstellar medium chemistry with neural operators

Galaxy formation and evolution critically depend on understanding the complex photo-chemical processes that govern the evolution and thermodynamics of the InterStellar Medium (ISM). Computationally, solving chemistry is among the most heavy tasks in cosmological and astrophysical simulations. The evolution of such non-equilibrium photo-chemical network relies on implicit, precise, computationally costly, ordinary differential equations (ODE) solvers. Here, we aim at substituting such procedural solvers with fast, pre-trained, emulators based on neural operators. We emulate a non-equilibrium chemical network up to H2 formation (9 species, 52 reactions) by adopting the DeepONet formalism, i.e. by splitting the ODE solver operator that maps the initial conditions and time evolution into a tensor product of two neural networks. We use textttKROME to generate a training set spanning -2≤ log(n/cm-3) ≤ 3.5, log(20) ≤log(T/K) ≤ 5.5, -6 ≤ log(ni/n) < 0, and by adopting an incident radiation field F sampled in 10 energy bins with a continuity prior. We separately train the solver for T and each ni for simeq 4.34,rm GPUhrs. Compared with the reference solutions obtained by textttKROME for single zone models, the typical precision obtained is of order 10-2, i.e. the 10 × better with a training that is 40 × less costly with respect to previous emulators which however considered only a fixed F. The present model achieves a speed-up of a factor of 128 × with respect to stiff ODE solvers. Our neural emulator represents a significant leap forward in the modeling of ISM chemistry, offering a good balance of precision, versatility, and computational efficiency.