Using cosmological simulations, we address the properties of high-redshift starforming galaxies (SFGs) across their main sequence (MS) in the plane of star-formation rate (SFR) versus stellar mass. We relate them to the evolution of galaxies through phases of gas compaction, depletion, possible replenishment, and eventual quenching. We find that the high-SFR galaxies in the upper envelope of the MS are compact, with high gas fractions and short depletion times ("blue nuggets"), while the lower-SFR galaxies in the lower envelope have lower central gas densities, lower gas fractions and longer depletion times, consistent with observed gradients across the MS. Stellarstructure gradients are negligible. The SFGs oscillate about the MS ridge on timescales ∼ 0.4 t Hubble (∼ 1 Gyr at z ∼ 3). The propagation upwards is due to gas compaction, triggered, e.g., by mergers, counter-rotating streams, and/or violent disc instabilities. The downturn at the upper envelope is due to central gas depletion by peak star formation and outflows while inflow from the shrunken gas disc is suppressed. An upturn at the lower envelope can occur once the extended disc has been replenished by fresh gas and a new compaction can be triggered, namely as long as the replenishment time is shorter than the depletion time. The mechanisms of gas compaction, depletion and replenishment confine the SFGs to the narrow (±0.3 dex) MS. Full quenching occurs in massive haloes (M vir > 10 11.5 M ) and/or at low redshifts (z < 3), where the replenishment time is long compared to the depletion time, explaining the observed bending down of the MS at the massive end.

Using cosmological simulations, we address the interplay between structure and star formation in high-redshift galaxies via the evolution of surface density profiles. Our sample consists of 26 galaxies evolving in the redshift range z = 7−1, spanning the stellar mass range (0.2−6.4)×10 10 M at z = 2. We recover the main trends by stacking the profiles in accordance to their evolution phases. Following a wet compaction event that typically occurs when the stellar mass is ∼ 10 9.5 M at z ∼ 2 − 4, the gas develops a cusp inside the effective radius, associated with a peak in star-formation rate (SFR). The SFR peak and the associated feedback, in the absence of further gas inflow to the centre, marks the onset of gas depletion from the central 1 kpc, leading to quenching of the central SFR. An extended, star-forming ring that forms by fresh gas during the central quenching process shows as a rising specific SFR (sSFR) profile, which is interpreted as inside-out quenching. Before quenching, the stellar density profile grows self-similarly, maintaining its log-log shape because the sSFR is similar at all radii. During the quenching process, the stellar density saturates to a constant value, especially in the inner 1 kpc. The stellar mass and SFR profiles deduced from observations show very similar shapes, consistent with the scenario of wet compaction leading to inside-out quenching and the subsequent saturation of a dense stellar core. We predict a cuspy gas profile during the blue nugget phase, and a gas-depleted core, sometimes surrounded by a ring, in the post-blue nugget phase.

The similarity between the distributions of spins for galaxies (λ gal ) and for dark-matter haloes (λ halo ), indicated both by simulations and observations, is naively interpreted as a one-to-one correlation between the spins of a galaxy and its host halo. This is used to predict galaxy sizes in semi-analytic models via R e ≃ λ halo R vir , where R e is the half-mass radius of the galaxy and R vir is the halo radius. Utilizing two different suites of zoom-in cosmological simulations, we find that λ gal and the λ halo of its host halo are in fact only barely correlated, especially at z 1. A general smearing of this correlation is expected based on the different spin histories, where the more recently accreted baryons through streams gain and then lose significant angular momentum compared to the gradually accumulated dark matter. Expecting the spins of baryons and dark matter to be correlated at accretion into R vir , the null correlation at the end reflects an anti-correlation between λ gal /λ halo and λ halo , which can partly arise from mergers and a compact star-forming phase that many galaxies undergo. On the other hand, the halo and galaxy spin vectors tend to be aligned, with a median cos θ = 0.6-0.7 between galaxy and halo, consistent with instreaming within a preferred plane. The galaxy spin is better correlated with the spin of the inner halo, but this largely reflects the effect of the baryons on the halo. Following the null spin correlation, λ halo is not a useful proxy for R e . While our simulations reproduce a general relation of the sort R e = AR vir , in agreement with observational estimates, the relation becomes tighter with A = 0.02(c/10) −0.7 , where c is the halo concentration, which in turn introduces a dependence on mass and redshift.

We study the evolution of global shapes of galaxies using cosmological simulations. The shapes refer to the components of dark matter (DM), stars and gas at the stellar half-mass radius. Most galaxies undergo a characteristic compaction event into a blue nugget at z ∼ 2 − 4, which marks the transition from a DM-dominated central body to a self-gravitating baryonic core. We find that in the high-z, DM-dominated phase, the stellar and DM systems tend to be triaxial, preferentially prolate and mutually aligned. The elongation is supported by an anisotropic velocity dispersion that originates from the assembly of the galaxy along a dominant large-scale filament. We estimate that torques by the dominant halo are capable of inducing the elongation of the stellar system and its alignment with the halo. Then, in association with the transition to self-gravity, small-pericenter orbits puff up and the DM and stellar systems evolve into a more spherical and oblate configuration, aligned with the gas disc and associated with rotation. This transition typically occurs when the stellar mass is ∼ 10 9 M and the escape velocity in the core is ∼ 100 km s −1 , indicating that supernova feedback may be effective in keeping the core DM-dominated and the system prolate. The early elongated phase itself may be responsible for the compaction event, and the transition to the oblate phase may be associated with the subsequent quenching in the core.

We use machine learning to identify in color images of high-redshift galaxies an astrophysical phenomenon predicted by cosmological simulations. This phenomenon, called the blue nugget (BN) phase, is the compact star-forming phase in the central regions of many growing galaxies that follows an earlier phase of gas compaction and is followed by a central quenching phase. We train a Convolutional Neural Network (CNN) with mock "observed" images of simulated galaxies at three phases of evolution: pre-BN, BN and post-BN, and demonstrate that the CNN successfully retrieves the three phases in other simulated galaxies. We show that BNs are identified by the CNN within a time window of ∼ 0.15 Hubble times. When the trained CNN is applied to observed galaxies from the CANDELS survey at z = 1 − 3, it successfully identifies galaxies at the three phases. We find that the observed BNs are preferentially found in galaxies at a characteristic stellar mass range, 10 9.2−10.3 M at all redshifts. This is consistent with the characteristic galaxy mass for BNs as detected in the simulations, and is meaningful because it is revealed in the observations when the direct information concerning the total Corresponding author: Marc Huertas-Company galaxy luminosity has been eliminated from the training set. This technique can be applied to the classification of other astrophysical phenomena for improved comparison of theory and observations in the era of large imaging surveys and cosmological simulations.