Scientists used deep learning to analyze quasar spectral data from SDSS-III; It successfully identified 107 rare neutral carbon sinks and provided new insights into the early evolution of galaxies. Their techniques could revolutionize the way astronomers use artificial intelligence to study the universe, complementing other research instruments such as the James Webb Space Telescope.

Recently, researchers searched for rare weak signals in quasar spectral data released by the Sloan Digital Sky Survey III (SDSS-III) program using deep learning neural networks. The team has presented a new method for studying the formation and evolution of galaxies, demonstrating the potential of artificial intelligence (AI) to detect rare, weak signals in large astronomical data sets. This recently published study Monthly Notices of the Royal Astronomical SocietyIt was conducted by an international team led by Professor Jian Ge from the Shanghai Astronomical Observatory of the Chinese Academy of Sciences.

Problems in identifying neutral carbon sinks

“Neutral carbon sinks” of cold gas and dust in the universe serve as key probes for studying the formation and evolution of galaxies. However, signals of neutral carbon absorption lines are weak and extremely rare. Astronomers have attempted to detect these absorbers in huge spectral datasets of quasars using traditional correlation methods.

“It’s like looking for a needle in a haystack,” Professor Ge said. In 2015, 66 neutral carbon sinks were detected in the spectra of tens of thousands of quasars previously released by SDSS; This was the largest number of samples ever obtained.

A breakthrough in absorber detection

In this study, Professor Ge’s team developed and trained deep neural networks with large numbers of simulated examples of neutral carbon absorption lines based on real observations. By applying these well-trained neural networks to SDSS-III data, the team identified 107 extremely rare neutral carbon sinks; This doubled the number of samples obtained in 2015 and found weaker signals than before.

Improving detection and understanding of galaxy evolution

By compiling the spectra of a large number of neutral carbon absorbers, the team greatly expanded the ability to detect a large number of different elements and directly measure metal losses from dust in the gas. The results showed that these first galaxies containing neutral carbon probes underwent rapid physical and chemical evolution when the universe was only three billion years old (the current age of the universe is 13.8 billion years). These galaxies entered a phase of evolution between the Large Magellanic Cloud (LMC) and the Milky Way (MW) and produced significant amounts of metal; some of these aggregated to form dust particles, leading to the observed dust reddening effect.

Additional findings from the James Webb Space Telescope

This discovery independently confirms recent findings from the James Webb Space Telescope (JWST), which detected diamond-like carbon dust in the universe’s oldest stars; This suggests that some galaxies are evolving much faster than previously expected, challenging existing models of galaxy formation and evolution.

Unlike JWST, which examines the emission spectra of galaxies, this study examines early galaxies by observing the absorption spectra of quasars. The application of well-trained neural networks to search for neutral carbon sinks provides a new tool for future studies of the early evolution of the Universe and galaxies, complementing JWST’s research methods.

Future directions and innovations of artificial intelligence

“There is a need to develop innovative artificial intelligence algorithms that can quickly, accurately and comprehensively discover rare and weak signals in massive astronomical data,” said Professor Ge. The team aims to improve the method presented in this work for image recognition by extracting multiple structures of interest to create artificial “multistructure” images for efficient training and detection of weak image signals.