Searching for Quiescent Galaxies over $3 < z < 6$ in JWST Surveys Using Manifold Learning

Abstract

Quiescent galaxies over $3<z<6$ are rare and puzzling. They formed and quenched within two billion years and simulations routinely struggle to predict their observed abundances. Developing a robust identification technique for these galaxies is crucial for constraining galaxy evolution models. Traditional rest-frame color-color selection techniques for quiescent galaxies are known to break down or require adjustments at $z\gtrsim3$. Recently, observed-frame color-color criteria have been established with JWST/NIRCam colors that efficiently pre-select high-redshift quiescent galaxies using only $\lesssim1\%$ of a given sample. In this work, the Uniform Manifold Approximation and Projection machine-learning technique is applied to pre-select quiescent galaxies over $3<z<6$ using observed NIRCam colors. From a parent sample of 43,926 galaxies in JADES, we ultimately find 44 quiescent candidates from a pool of $\approx2,300$ galaxies. This is about five times fewer galaxies than what would be pre-selected using color-color criteria. Two-thirds of these candidates can be pre-selected from a pool as small as 247, which is about twice as efficient as existing observed-frame color selection techniques. Nearly two-thirds of the candidates are new discoveries and include quiescent galaxies with mass-weighted ages as young as $\lesssim300$ Myr. We obtain number densities in agreement with the literature at $z<4$ and find generally higher abundances at $z>4$, although our measurements are consistent within errors. This technique may be applied to other JWST surveys.