# A fast neural emulator for interstellar chemistry

**Authors:** A. Asensio Ramos, C. Westendorp Plaza, D. Navarro-Almaida, P. Rivière-Marichalar, V. Wakelam, A. Fuente  
**arXiv:** [2406.02387](https://arxiv.org/abs/2406.02387)  
**Date:** 2024-06-04  
**Summary page:** https://www.summarizepaper.com/en/arxiv-id/2406.02387v1/

## Key points

- Numerical simulations are crucial for studying the evolution of interstellar matter during star formation
- Radiation-magnetohydrodynamical (R-MHD) simulations are state-of-the-art for analyzing gravitational collapse in star formation
- Magnetic fields play a significant role in processes like collapse, fragmentation, and disk formation
- Virtual particles are used to store essential parameters for chemistry computation in complex astrophysical environments
- Fast chemistry emulators based on neural networks have been developed to enhance predictive capabilities in astrochemical modeling

## Summary

, , , , 


In recent years, numerical simulations have become crucial tools for studying the evolution of interstellar matter during the star formation process. Various approaches have been employed, including analytical one-dimensional models, two-dimensional semi-analytical models, and three-dimensional radiation-magnetohydrodynamical (R-MHD) simulations. Among these, R-MHD simulations have emerged as the state-of-the-art method for analyzing the gravitational collapse of prestellar cores into protostellar objects. This highlights the significant role of magnetic fields in processes such as collapse, fragmentation, and disk formation. While some advanced simulations incorporate a reduced chemical network to provide key parameters for MHD calculations, fully coupling MHD evolution with a comprehensive chemical reaction network remains challenging due to computational limitations. As a feasible alternative, virtual particles have been utilized to store essential parameters for chemistry computation in complex astrophysical environments. However, these calculations are computationally expensive and hinder a comprehensive comparison of 3D chemo-MHD simulations with observations. , , and  play vital roles in understanding the interstellar medium by employing chemical models consisting of coupled ordinary differential equations representing a chemical network. These models require input parameters such as gas physical conditions, cosmic ray ionization rate per H atom, incident UV field, visual extinction, grain properties, initial chemical composition, and elemental abundances. However, computing large grids of complex chemical models across various astrophysical environments is prohibitively time-consuming. To address these challenges and enhance predictive capabilities in astrochemical modeling,< kd>Fast chemistry emulators</ kd> have been developed using neural networks. These emulators enable rapid calculations based on real astrochemical models while maintaining high accuracy levels. In this study, a fast neural emulator based on conditional neural fields was presented for the astrochemical code Nautilus. The model efficiently predicts the abundance of 192 species over arbitrary timescales between 1 and 10^7 years with uncertainties below 0.2 dex for all species. The implementation of fast chemistry emulators opens up new possibilities for conducting more intricate forward models to gain deeper insights into the physical properties of the interstellar medium. By leveraging these advanced tools, researchers can explore complex relationships within astrochemical systems and uncover critical factors influencing phenomena such as electron density variations in low-density gases due to changes in initial sulfur abundance. This enhanced understanding has implications for shaping our knowledge of  within interstellar environments and paves the way for further advancements in astrophysical research.