Neural Deprojection of Galaxy Stellar Mass Profiles

TL;DR

This work presents a neural deprojection that maps Nuker model parameters to an axisymmetric MGE representation, enabling physically realistic stellar mass profiles without optical imaging. By embedding this mapping in the differentiable dynamical pipeline SuperMAGE, the authors perform Bayesian SMBH mass inference directly from ALMA radio-visibilities, demonstrating results consistent with optical MGE-based analyses while extending applicability to dust-obscured and AGN-dominated galaxies. The approach yields a tighter, more robust mass model and opens the path for SMBH measurements in challenging regimes, including high-redshift gravitationally lensed galaxies. Overall, the combination of neural deprojection, differentiable dynamical modelling, and radio-velocity data offers a powerful framework for precise SMBH mass measurements beyond the limitations of optical imaging.

Abstract

We introduce a neural approach to dynamical modeling of galaxies that replaces traditional imaging-based deprojections with a differentiable mapping. Specifically, we train a neural network to translate Nuker profile parameters into analytically deprojectable Multi Gaussian Expansion components, enabling physically realistic stellar mass models without requiring optical observations. We integrate this model into SuperMAGE, a differentiable dynamical modelling pipeline for Bayesian inference of supermassive black hole masses. Applied to ALMA data, our approach finds results consistent with state-of-the-art models while extending applicability to dust-obscured and active galaxies where optical data analysis is challenging.

Figures (4)

• Figure 1: A flowchart that illustrates the architecture of the NN that maps Nuker parameters to MGE parameters.
• Figure 2: Credible intervals for the black hole mass in NGC4697 (box: $1\sigma$, whiskers: $3\sigma$). To produce self-consistent results, KinMS requires an empirically-derived uncertainty rescaling to account for systematic uncertainty (this rescaling has been applied to the second and third box plots). The first two box plots show the KinMS model's inferred SMBH mass given a stellar light profile with the innermost Gaussian removed to account for the presence of light emitted by the AGN ("w/o AGN"); the third box plot shows the KinMS model's inferred black hole mass if the innermost Gaussian is kept as part of the stellar light profile ("w/ AGN").
• Figure 3: Left: 2D projected mass density for the optically-fitted MGE profile and our dynamically-fitted Nuker profile. Right: Derived orbital velocity curves for the KinMS+MGE model and our SuperMAGE+Nuker model. The credible intervals for KinMS include the uncertainty rescaling.
• Figure 4: Corner plot for posterior samples for all free parameters in our SuperMAGE+Nuker model. Note that the break radius is constrained to values outside the maximum extent of gas in the galaxy (roughly 3 arcseconds), which indicates that the galaxy is well described by a single power law.