Neural network-based emulation of interstellar medium models

Abstract

The interpretation of observations of atomic and molecular tracers in the galactic and extragalactic interstellar medium (ISM) requires comparisons with state-of-the-art astrophysical models to infer some physical conditions. Usually, ISM models are too time-consuming for such inference procedures, as they call for numerous model evaluations. As a result, they are often replaced by an interpolation of a grid of precomputed models. We propose a new general method to derive faster, lighter, and more accurate approximations of the model from a grid of precomputed models. These emulators are defined with artificial neural networks (ANNs) designed and trained to address the specificities inherent in ISM models. Indeed, such models often predict many observables (e.g., line intensities) from just a few input physical parameters and can yield outliers due to numerical instabilities or physical bistabilities. We propose applying five strategies to address these characteristics: 1) an outlier removal procedure; 2) a clustering method that yields homogeneous subsets of lines that are simpler to predict with different ANNs; 3) a dimension reduction technique that enables to adequately size the network architecture; 4) the physical inputs are augmented with a polynomial transform to ease the learning of nonlinearities; and 5) a dense architecture to ease the learning of simple relations. We compare the proposed ANNs with standard classes of interpolation methods to emulate the Meudon PDR code, a representative ISM numerical model. Combinations of the proposed strategies outperform all interpolation methods by a factor of 2 on the average error, reaching 4.5% on the Meudon PDR code. These networks are also 1000 times faster than accurate interpolation methods and require ten to forty times less memory. This work will enable efficient inferences on wide-field multiline observations of the ISM.

References in corpus (18)

• An Analysis of the Shapes of Interstellar Extinction Curves. V. The IR-Through-UV Curve Morphology
• H3+ in Diffuse Interstellar Clouds: a Tracer for the Cosmic-Ray Ionization Rate
• HI-to-H2 Transitions and H I Column Densities in Galaxy Star-Forming Regions
• The anatomy of the Orion B Giant Molecular Cloud: A local template for studies of nearby galaxies
• Compression and ablation of the photo-irradiated cloud the Orion Bar
• UCLCHEM: A Gas-Grain Chemical Code
• Surveying the reach and maturity of machine learning and artificial intelligence in astronomy
• A New Reference Chemical Composition for TMC-1
• The penetration of FUV radiation into molecular clouds
• Models of irradiated molecular shocks
• Tracing Interstellar Heating: An ALCHEMI Measurement of the HCN Isomers in NGC 253
• Inferring the HII region escape fraction of ionizing photons from infrared emission lines in metal-poor star-forming dwarf galaxies
• Chemulator: Fast, accurate thermochemistry for dynamical models through emulation
• ROBO: a Model and a Code for the Study of the Interstellar Medium
• Deep learning denoising by dimension reduction: Application to the ORION-B line cubes
• Machine learning-accelerated chemistry modeling of protoplanetary disks
• Understanding the Formation and Evolution of Interstellar Ices: A Bayesian Approach
• UCLCHEMCMC: A MCMC Inference tool for Physical Parameters of Molecular Clouds