Abstract:The study of galaxy formation and evolution critically depends on our understanding of the complex photo-chemical processes that govern the evolution and thermodynamics of the interstellar medium (ISM). In a computational sense, resolving the chemistry is among the weightiest tasks in cosmological and astrophysical simulations. Astrophysical simulations can include photo-chemical models that allow for a wide range of densities ($n$), abundances of different species ($n_i/n$) and temperature ($T$), and plausible evolution scenarios of the ISM under the action of a radiation field ($ F $) with different spectral shapes and intensities. The evolution of such a non-equilibrium photo-chemical network relies on implicit, precise, computationally costly, ordinary differential equations (ODE) solvers. Here, we aim to substitute such procedural solvers with fast, pre-trained emulators based on neural operators. We emulated a non-equilibrium chemical network up to H$_2$ formation (9 species, 52 reactions) by adopting the DeepONet formalism, namely: by splitting the ODE solver operator that maps the initial conditions and time evolution into a tensor product of two neural networks (named branch and trunk). We used $ KROME $ to generate a training set, spanning $-2 K < 0$, and adopting an incident radiation field, $ F $ sampled in ten energy bins with a continuity prior. We separately trained the solver for $T$ and each $n_i$ for $ GPUhrs$. Compared with the reference solutions obtained by $ KROME $ for single-zone models, the typical precision obtained is of the order of $10^ $, that is, it is 10 times better when using a training that is 40 times less costly, with respect to previous emulators that only considered a fixed $ F $. DeepONet also performs well for $T$ and $n_i$ outside the range of the training sample. Furthermore, the emulator aptly reproduces the ion and temperature profiles of photo dissociation regions as well; namely, by giving errors that are comparable to the typical difference between various photo-ionization codes. 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 modelling of ISM chemistry, offering a good balance of precision, versatility, and computational efficiency. Nevertheless, further work is required to address the challenges represented by the extrapolation beyond the training time domain and the removal of potential outliers.

Cue: A Fast and Flexible Photoionization Emulator for Modeling Nebular Emission Powered By Almost Any Ionizing Source

Emulation of the Cosmic Dawn 21-cm Power Spectrum and Classification of Excess Radio Models Using an Artificial Neural Network

21cmEMU: an emulator of 21cmFAST summary observables

A neural network emulator for the Lyman-$α$ 1D flux power spectrum

Emulating the interstellar medium chemistry with neural operators

Emulating Radiative Transfer with Artificial Neural Networks

deepCool: Fast and Accurate Estimation of Cooling Rates in Irradiated Gas with Artificial Neural Networks

Emulating radiation transport on cosmological scale using a denoising Unet

Introducing PyCross: PyCloudy Rendering Of Shape Software for pseudo 3D ionisation modelling of nebulae

CIGALE: a python Code Investigating GALaxy Emission

A fast neural emulator for interstellar chemistry

Photoionization Models for the Semi-Forbidden C III] 1909 Emission in Star-Forming Galaxies

Semi-supervised Learning for Detecting Inverse Compton Emission in Galaxy Clusters

Emission line predictions for mock galaxy catalogues: a new differentiable and empirical mapping from DESI

Using [Ne V]/[Ne III] to Understand the Nature of Extreme-Ionization Galaxies

A New Photoionization Model of the Narrow Line Region in Active Galactic Nuclei

Photoionization analysis of chemodynamical dwarf galaxies simulations. II. Detailed calculation of diffuse ionizing radiation

Neural Network Constraints on the Cosmic-Ray Ionization Rate and Other Physical Conditions in NGC 253 with ALCHEMI Measurements of HCN and HNC

Kilonova Light-Curve Interpolation with Neural Networks