Kinematic Distortions of the High-Redshift Universe as Seen from Quasar Proper Motions

TL;DR

This paper uses machine learning and Gaia proper motions to detect potential kinematic distortions in the high-redshift universe, providing a new observational test for cosmological models.

Contribution

It introduces a novel method combining neural network redshift predictions with vector spherical harmonic analysis of Gaia data to identify redshift-dependent kinematic patterns.

Findings

Detected significant differences in proper motion patterns across redshift bins.

Identified specific harmonic signals such as a rigid spin and dipole distortions.

Results suggest possible systematic errors or new cosmological effects.

Abstract

Advances in optical astrometry allow us to infer the non-radial kinematic structure of the Universe directly from observations. Here I use a supervised machine learning neural network method to predict 1.57 million redshifts based on several photometric and metadata classifier parameters from the unWISE mid-infrared database and from Gaia. These estimates are used to divide the sample into three redshift bins: 1-2, 2-3, and $>3$. For each subset, all available Gaia proper motions are used in a global vector spherical harmonic solution to degree 3 (30 fitting vector functions). I find significant differences in a few fitted proper motion patterns at different redshifts. The largest signals are seen in the comparison of the vector spherical harmonic fits for the 1-2 and 2-3 redshift bins. The significant harmonics include a rigid spin, a dipole glide from the north Galactic pole to the south and an additional quadrupole distortion. Validation tests with filtered subsamples indicate that the detected effect can be caused by hidden systematic errors in astrometry. The results are verified by using an independent source of redshifts and computing the observer's Galactocentric acceleration. This study offers a new observational test of alternative cosmological models.