Galaxy spectroscopy without spectra: Galaxy properties from photometric images with conditional diffusion models

TL;DR

This work is the first attempt in the literature to infer velocity dispersion from photometric images, and can predict the presence of an active galactic nucleus up to an accuracy of 82%.

Abstract

Modern spectroscopic surveys can only target a small fraction of the vast amount of photometrically cataloged sources in wide-field surveys. Here, we report the development of a generative AI method capable of predicting optical galaxy spectra from photometric broad-band images alone. This method draws from the latest advances in diffusion models in combination with contrastive networks. We pass multi-band galaxy images into the architecture to obtain optical spectra. From these, robust values for galaxy properties can be derived with any methods in the spectroscopic toolbox, such as standard population synthesis techniques and Lick indices. When trained and tested on 64x64-pixel images from the Sloan Digital Sky Survey, the global bimodality of star-forming and quiescent galaxies in photometric space is recovered, as well as a mass-metallicity relation of star-forming galaxies. The comparison between the observed and the artificially created spectra shows good agreement in overall metallicity, age, Dn4000, stellar velocity dispersion, and E(B-V) values. Photometric redshift estimates of our generative algorithm can compete with other current, specialized deep-learning techniques. Moreover, this work is the first attempt in the literature to infer velocity dispersion from photometric images. Additionally, we can predict the presence of an active galactic nucleus up to an accuracy of 82%. With our method, scientifically interesting galaxy properties, normally requiring spectroscopic inputs, can be obtained in future data sets from large-scale photometric surveys alone. The spectra prediction via AI can further assist in creating realistic mock catalogs.

Figures (15)

• Figure 1: Visualization of photometric images of different galaxies in the training set, which serves as the input for the machine learning algorithm. The corresponding galaxy name can be found at the top of the image and the depicted band is written on each image. We chose to visualize diverse objects to better emphasize the variety of galaxy images in the training set.
• Figure 2: Visualization of our pipeline. We use a conditional diffusion model to generate candidate low-resolution spectra for the given image and select the best matching candidates with a contrastive network. Then, we use a conditional diffusion model to upsample the candidate spectra to full-resolution candidates and select the final predicted spectrum with the second contrastive network.
• Figure 3: Histogram of $\overline{\Delta}$, Mean Square Error (MSE) and $\chi^2/N$ (with N the number of flux-points) of all galaxies in the test set split into different morphologies. When taking into account the uncertainty of the observed spectra, all three groups perform equally well. The morphological categories "elliptical", "spiral", and "uncertain" were determined by the citizen science project Galaxy Zoo Lintott2008Lintott2011). Galaxies were labeled as "uncertain" if their images were not clearly voted on as spiral or elliptical. Note that this does not say much about their spectral classification as, for instance, a post-merger starburst galaxy could belong to this group.
• Figure 4: Comparison between the spectroscopic redshift from the observed spectra (z$_{\textrm{spec}}$) and the prediction from our generative AI for the galaxies (z$_{\textrm{predict}}$) in the test set. Instead of single data points, the overall kernel density estimate is shown for a clearer point of view. The grey dotted lines mark the regions with catastrophic outliers with $|\Delta z /(1+z_{spec})| >0.05$. The black dotted line shows the one-to-one relation.
• Figure 5: Comparison of Lick Indices measured using the observed spectra and the predicted spectra. Spiral galaxies were excluded. The color coding of the density plots indicates how many galaxies reside at a specific (x,y)-position and is kept frozen in each row for comparability. Pearson correlation coefficients $\rho$ are printed in each subfigure. Generally, the coefficient’s value ranges from +1 (perfect positive correlation) to -1 (perfect negative correlation), with 0 indicating no correlation.