The Evolutionary Pathways of Disk-, Bulge-, and Halo-dominated Galaxies

Cosmologically-motivated models suggest that stars tend to conserve their binding energy during galaxy mergers. Ex-situ stars, thus, can be loosely bound and can hence populate galaxies in a broad range of galactocentric distances (e.g. Barnes, 1988; Hopkins et al., 2009; Amorisco, 2017). The constituent stars of stellar halos are fossil records of the hierarchical merging process (e.g. Deason et al., 2016; D’Souza & Bell, 2018; Monachesi et al., 2019): mergers with larger satellites produce more massive, higher-metallicity stellar halos, and can thus reproduce the recently observed stellar halo metallicity-mass relation (discovered by an HST imaging survey of nearby galaxies, GHOSTS, Harmsen et al., 2017). Studies of the hierarchical growth of structures have reached similar conclusions using large-scale, hydrodynamic cosmological simulations (e.g. Illustris Pillepich et al., 2014; Rodriguez-Gomez et al., 2016; Pop et al., 2018).

It is often argued that massive, compact classical bulges formed at early cosmic epochs via various pathways such as early gas-rich accretions, violent disk instabilities, or misaligned inflows of gas. Such classical bulges, thus, are largely composed of stars formed in-situ and characterized by low binding energy. Bell et al. (2017) showed that galaxies with massive classical bulges have diverse merger histories, and no clear correlation with properties of the stellar halos has been found. It is, thus, plausible that bulges are indeed dominated by in-situ chaotic processes. The classical conception that bulges are produced by mergers may therefore not hold in all cases.

Figures 2 and 2 show the distributions of the morphology and relative importance of rotation for a selection of TNG50 galaxies when viewed edge-on. They are characterized by the mass fraction of kinematic bulge ($f_{\rm b}$, $y$-axis) and halo ($f_{\rm h}$, $x$-axis) derived by auto-GMM. We normalize the surface density map of each galaxy by its maximum value to gain equal contrast for galaxies with different stellar masses. Obviously, disk galaxies have strong rotation, thus located at the bottom-left corner. Elliptical galaxies mainly lie above the dashed line where the overall mass fraction of spheroids $f_{\rm sph}=f_{\rm b}+f_{\rm h}$ is larger than 0.5. However, galaxies with massive bulges seem to be not clearly distinguishable from those having massive halos in observations, even when taking kinematics into account. This issue is the more serious the lower the inclination. Therefore, a severe degeneracy exists in classical morphological decompositions of bulge vs. halo stars, even though the central massive concentration indeed becomes more prominent with the increase of bulge mass fraction.

As discussed in Section 3, the stars of bulges and stellar halos defined by our kinematic method are separated by their binding energies, which is consistent with our physical expectation. Both bulges and halos have similarly weak rotation; however, as shown by the spatial distribution of their stellar particles in Figure 4 and 1D surface density profiles in Figure 4, bulge stars (yellow) are tightly bound around the galactic central regions. Halo stars (red) are loosely bound, comprising the diffuse envelopes. Halo stars, that move on highly elliptical orbits, are able to pass through the central regions that are dominated by bulge stars. The half-mass radii of bulges are generally less than 2 kpc, while those of the stellar halos vary in a broad range of 2-10 kpc.

The fact that stellar halos approximately follow the S$\acute{e}$rsic function, see Figure 4, may induce a serious difficulty in making accurate morphological decompositions of galaxies, and hence in advancing interpretations for their formation processes. In order to illustrate this issue clearly, we take the famous Sombrero Galaxy (M104/NGC 4594, see Figure 5) as an example. The Sombrero Galaxy is regarded as one of the most unusual galaxy having a disk embedded in an extremely large “bulge”. Gadotti & Sánchez-Janssen (2012) argued that the bulge mass fraction can be reduced from $77\%$ to $<10\%$ in the Sombrero Galaxy if an outer spheroidal component, i.e., a stellar halo, is considered. We highlight four Sombrero visual analogues with red squares in Figures 2 and 2. As we can see in Figures 4 and 4, the huge “bulges” of Sombrero analogues are largely contributed by kinematic halos. The mass fraction of their bulges can vary from 0 to 0.5. Idealized synthetic images of Sombrero-like galaxies from TNG50 are shown too in Figure 5, following the procedure described in Rodriguez-Gomez et al. (2019).

auto-GMM allows us to break the degeneracy between bulges and halos even in the central regions of galaxies. In this picture, kinematic bulges qualitatively correspond to classical bulges. Normal elliptical galaxies are largely dominated by kinematic stellar halos, which challenges the general idea that elliptical galaxies (i.e., slow rotator ETGs) are the same objects as classical bulges in disk galaxies, obeying the Kormendy relation (e.g. Kormendy, 1977; Gadotti, 2009) and the $M_{\rm bh}-\sigma_{\rm s}$ relation (e.g. Kormendy & Ho, 2013). About 52 per cent of our central elliptical galaxies, that are selected by the spheroidal mass fraction $f_{\rm sph}\geq 0.5$, are halo-dominated galaxies. It is worth emphasizing that although stellar halos have generally much lower surface density than other structures on the midplane, their overall mass fractions can be large due to their wide extent reaching tens, if not hundreds, of kpc distance.

It is expected that intrinsic structures in galaxies are reflected by their morphological and kinematic properties, but possibly in a non-linear way. In this section, we discuss the relation between galaxies classified by our kinematic method and their global morphological and kinematic properties.

Figure 6 shows some basic properties: stellar half-mass radius $r_{\rm e}$, global rotation $K_{\rm rot}$ (Sales et al., 2010), star formation rate (SFR hereafter), and the mass fractions of kinematic structures (see Figure 27 for TNG100 galaxies in appendix A). $K_{\rm rot}=\langle v_{\phi}^{2}/v^{2}\rangle$, where $v_{\phi}$ and $v$ are the cylindrical rotation velocity and total velocity, respectively, quantifies the relative importance of cylindrical rotation. Each data point is coloured by the mass fraction of its spheroid, bulge, and halo, respectively, from top to bottom. Clearly, galaxies with different kinematic structures have very different properties. This is a confirmation of the bounty of both the galaxy formation model underlying TNG50 as well as our kinematically-motivated stellar decomposition method. The galaxies that are dominated by spheroidal structures (red points in the top panels) generally have relatively compact morphologies, quiescent SF and weak rotation. The blue points, mainly corresponding to galaxies dominated by kinematic disks, are generally extended galaxies with active SF and strong rotation, that preferentially populate the low-mass end ($M_{\rm s}<10^{10.6}M_{\odot}$) of the distribution.

It is clearly shown in the top-left panel of Figure 6 that galaxies dominated by spheroidal structures are common in massive galaxies producing the well-known mass-size relation, where disk galaxies are rare. Systematic comparisons between the mass-size relation in observations and that in IllustrisTNG are given in Rodriguez-Gomez et al. (2019); Genel et al. (2018) for TNG100 and in Pillepich et al. (2019) for TNG50. As shown in the middle panels of Figure 6, galaxies with more massive bulges are generally more compact (smaller size and larger central density), while those with massive halos (bottom panels) are not that dramatically different in $r_{\rm e}$ from galaxies dominated by disky structures. Interestingly, we can see that many galaxies with massive stellar halos are as extended as disk galaxies in less massive cases of $M_{\rm s}\lesssim 10^{10.6}$. Galaxies with massive bulges are generally the most compact objects over a broad mass range.

The mass fraction of spheroidal components decomposed by auto-GMM is tightly correlated with $K_{\rm rot}$ that is almost independent of galaxy stellar mass. $K_{\rm rot}>0.5$ has been widely used as a criterion to select disk galaxies in simulations. This criterion selects almost the same group of galaxies as using $f_{\rm sph}<0.5$. Both galaxies with massive bulges and halos are thus generally classified as elliptical/early-type galaxies, while they are clearly different types of galaxies. The galaxies with massive bulges have somewhat stronger rotation ($K_{\rm rot}\sim 0.4-0.6$) and more disky morphology (Figure 2) than those with massive halos. This suggests that galaxies with massive bulges are analogues of fast rotator ETGs from both morphological and kinematic points of view. But there is no clear dividing line in $K_{\rm rot}$ that can separate them from galaxies with massive halos that are slow rotator analogues.

It is worth mentioning that, at the low-mass end, many galaxies dominated by spheroidal components are still actively forming stars, falling on the main sequence of disk galaxies (blue dots in the right panels of Figure 6). This result suggests the quenching is unlikely to be directly correlated with the growth of either bulges or halos in central galaxies. It is worth emphasizing that many star-forming galaxies with massive spheroids are Sombrero analogues in observations, but cannot be distinguished easily from spiral galaxies. Disks in such galaxies generally still have at least 20 per cent of their total stellar masses. This fraction may be even larger using the classical bulge-disk decomposition in morphology due to the contamination of stellar halos in low-inclination cases, as suggested by Du et al. (2020). Nevertheless, spiral structures are commonly visible in the face-on view. Sombrero analogues are, thus, likely to be classified as disk galaxies, instead of elliptical ones. A systematic study is required to make a robust conclusion on whether SFRs are less sufficiently suppressed in elliptical galaxies from the IllustrisTNG simulations.

Figure 7 shows the mass fractions of kinematic structures for all central galaxies from TNG50 in the $10^{10-11.5}\ M_{\odot}$ stellar mass range. The ratios of stellar halo, bulge, and disk mass, respectively, to the total stellar mass are denoted with $f_{\rm h}$, $f_{\rm b}$, and $f_{\rm d}$. In the left panel of Figure 7, we select three groups of galaxies:

• •

  Disk-dominated 1 and 2: $f_{\rm h}<0.2$. For the groups 1 (blue rectangle) and 2 (green rectangle), $f_{\rm b}$ is $<0.1$ and $0.1-0.2$, respectively.

• •

  Bulge-dominated: $f_{\rm b}\geq 0.2$ and $f_{\rm h}<0.2$, orange rectangle.

• •

  Halo-dominated: $f_{\rm b}<0.2$ and $f_{\rm h}\geq 0.4$, red rectangle.

From left to right in the right panel, the mass fraction of disky structures increases, thus galaxies change from spheroidal early-type to disky late-type ones. A large group of galaxies dominated by disks clusters at the bottom-right corner in the right panel of Figure 7 (also the bottom-left corner in the left panel). These galaxies are akin to pure-disk/bulgeless galaxies in observations (see their edge-on view in Figure 2). Note that the mass fraction of disks $f_{\rm d}$ is obtained by summing all stars in their cold and warm disks (which includes any disky/pseudo bulge). The mass fraction decrease of disky structures leads to the increase of either a bulge or a halo, thus two branches. Galaxies on the lower branch have relatively more massive halos.

Figure 8 shows the fraction of each galaxy group as a function of stellar mass. It is clear that the relatively low-mass galaxies (stellar mass $<10^{10.6}M_{\odot}$) in TNG100 have much more bulge-dominated cases. Star formation is generally still active in such galaxies. It is thus consistent with the conclusion of Rodriguez-Gomez et al. (2019) that TNG100 produces many blue spheroids. Similarly, Du et al. (2020) showed that disk galaxies in TNG100 generate a dramatically lower fraction of cold disks than those extracted in CALIFA galaxies (Zhu et al., 2018a). The result in this paper suggests that this issue has been significantly improved in TNG50, if not completely solved, possibly due to its better resolution.

In total, among the 541 central TNG50 galaxies, 183 ($\sim 34\%$) are disk-dominated. This fraction drops to 14% (75 galaxies) if we only count the galaxies that are dominated by cold disks (mass fraction $f_{\rm cd}\geq 0.5$), which is a conservative definition of disk-dominated galaxies. About 1% (5 galaxies) have cold disks with $f_{\rm cd}\geq 0.7$. The conclusions in Section 6 will not be significantly affected if we distinguish bulge- and disk-dominated galaxies according to the mass fraction of their cold disks.

Dissipationless “dry” mergers in the later phase are expected to be destructive for disky structures. In Figure 7, individual simulated galaxies are color-coded by the amount of stellar mass that was not formed in the galaxies lying along the main progenitor branch in the merger trees, i.e., ex-situ, estimated using the method of Rodriguez-Gomez et al. (2016). It is clear that the mass fraction of ex-situ stars is generally $<0.1$ in both disk- and bulge-dominated galaxies, while it is much larger in halo-dominated galaxies.

In the upper panels of Figure 9, we show the merger frequency of galaxies with different kinematic structures. It measures the fraction of galaxies that have experienced at least one merger of certain mass ratio since a particular redshift. The lower panels show number counts of galaxies. Clearly, the mass fraction of a kinematic stellar halo has a strong positive correlation with mergers. About 80 per cent of the central galaxies with massive stellar halos of mass fraction $f_{\rm h}>0.4$ have experienced at least one merger of stellar mass ratio $\geq 0.1$ since $z=1$ (see blue curves for TNG50 and gray histograms for TNG100). More than 50 per cent of such galaxies are associated with 0.25 major mergers (red profiles for TNG50 and black histogram for TNG100), in agreement with Penoyre et al. (2017) and Lagos et al. (2018).

Despite the somewhat artificial classification, both disk- and bulge-dominated galaxies have rarely been affected by mergers in the past 10 Gyr ($z<2$), thus they are likely to be two fundamentally-different types of galaxies in comparison to halo-dominated ones. Mergers have been infrequent in the late-phase of cosmic evolution for bulge-dominated galaxies, so that such bulges must have been generated by either internal processes or in an earlier-phase of the Universe. Bulge- and disk-dominated galaxies may be the two ends of a continuous distribution. The properties of such galaxies are able to record important information of their “initial conditions” after the early phase, which is likely lost in mergers. The formation of halo-dominated galaxies, in contrast, is tightly correlated with mergers that happen late, thus are relatively dry. They may form via major mergers between the two fundamental types of galaxies, or via ex-situ stellar accumulation through minor mergers with low-mass satellites. In order to gain new insights into galaxy properties discussed here and their evolutionary histories, we trace different types of galaxies back to high redshifts in the next two sections.

Figure 16 shows the extended and compact pathways on the mass-size diagram. The color represents the mass fraction of bulges in both disk- and bulge-dominated galaxies at $z=0$; the gray data points correspond to their progenitors at $z=1.5$. Disk-dominated galaxies generally follow the extended pathway, highlighted by the blue arrow. Bulge-dominated galaxies follow the compact pathway that has two phases: (1) a compact phase during which the mass grows significantly while the size changes little, thus forming bulges; (2) the size increases significantly while the mass grows relatively little, thus building up disky structures. Massive cases with $M_{\rm s}\gtrsim 10^{10.3}\ M_{\odot}$ have almost passed through the phase (1) at $z\sim 1.0$, while the compact phase seems to last to $z=0$ for less massive cases.

The progenitors of massive bulge-dominated galaxies (linked by the dashed-gray lines) with $M_{\rm s}\gtrsim 10^{10.5}\ M_{\odot}$ have already been rather massive and compact at $z=1.5$. They then evolve into the phase (2) of the compact pathway, during which diffuse disk structures are assembled gradually, thus their sizes increase in a similar way to disk-dominated ones. Figures 17 and 18 show two prototypes of massive disk- and bulge-dominated galaxies, named D1 and B1 (marked by squares in Figure 16), respectively, to illustrate the dramatically different evolutionary pathways between them. Stellar particles are classified into the structure that has the largest likelihood at $z=0$ via applying our kinematic decomposition algorithm. From top to bottom, we show the surface density maps in both face-on and edge-on views for total, cold disk, warm disk, bulge, and halo.

Three quantities (red words in the top and bottom textboxes of Figures 17 and 18) are used to characterize the number fractions of stars that originate from an earlier-phase evolution, external/ex-situ mergers, and internal/in-situ SF (from top to bottom), respectively, for each kinematic structure (see details of their definitions in the footnote⁴⁴4$N_{i,z}$ is the total stellar particle number of a certain structure $i$ at redshift $z$. Tracing back from $z_{1}$ to an earlier time point $z_{2}$, the new stellar particle members of each structure during $z_{1}-z_{2}$ are classified into two origins: ex-situ accretion and SF, i.e., in-situ. The ex-situ part is estimated by the stars that are not belong to this galaxy at $z_{2}$; the in-situ part is newly formed stars in a time span between two snapshots from $z_{2}$ to $z_{1}$.). For example, the kinematic cold disk of D1 contributes $54.4\%$ of its total stellar mass at $z=0$. At $z=1.5$, only $3.5\%$ of cold disk stars, i.e., $3.5\%N_{i,0}$, have already existed in this galaxy, where $N_{i,0}$ is the total stellar particle number of a certain structure $i$ that is cold disk here. During $z=1.5-1.0$, it increases by $\approx 11.0\%N_{i,0}$, where SF and ex-situ accretion contribute $10.97\%N_{i,0}$ and $0\%N_{i,0}$, respectively.

Clearly, the properties at $z=1.5$ are largely dominated by their early-phase evolution. At $z=1.5$, the B1 object (Figure 18) has already assembled 53.8 percent of its stars found at $z=0$, while D1 (Figure 17) has only had $20.9\%$ of its stars. The half-mass radius of B1, marked by dashed circles, is dramatically smaller than that of D1. A massive central concentration, i.e., bulge, is clearly visible in B1. The overall properties of the kinematic bulge in B1 changes mildly since $z=1.5$. Without experiencing mergers, the halo masses of both D1 and B1 also change little.

During $z=0-1.5$, an extended cold disk forms gradually in both D1 and B1, which leads to the increase of their galaxy sizes. The growth of cold disks coincides well with the SF shown in Figure 19. At $z<1$, B1 is gradually quenched inside-out (see Figures 15 and 15 for statistical results). An extended SF ring that is also gas-rich is formed. Zolotov et al. (2015) suggested that such a ring is a natural result of the accretion of cold gas with high angular momentum from the cosmic web into this node. Moreover, Dekel et al. (2020) showed that the existence of a massive central concentration, i.e., bulge, can suppress inward gas transport, which possibly also gives rise to a SF ring. In conclusion, the existence of a massive bulge formed in the early phase evolution makes bulge-dominated galaxies more compact than disk-dominated ones. Both of them are able to generate similar disky structures by internal SF in the later phase. Massive bulge-dominated galaxies, however, are likely to be quenched at low redshifts, thus classified as fast-rotator ETGs with massive bulges.

There is a dramatic difference in the size growth of disk- and bulge-dominated galaxies with $M_{\rm s}\lesssim 10^{10.3}\ M_{\odot}$, shown in Figures 15 and 16. The sizes of bulge-dominated galaxies are nearly flat or even decrease slightly while their stellar masses grow fast at $z\gtrsim 1$, while the $r_{\rm e}$ of disk-dominated galaxies increases significantly by a factor of $\sim 2.5$. However, it is surprising that the overall SFRs of such bulge- and disk-dominated galaxies follow a similar trend, though bulge-dominated galaxies have slightly higher SFR at $z>1$, but lower at $z<1$, than disk-dominated ones. The difference in SFR is not as significant as that in size between disk- and bulge-dominated galaxies. Therefore, the difference of compact and extended pathways must be due to the spatial distribution of their SF. Considering that both bulge- and disk-dominated galaxies are weakly affected by mergers, we speculate that their SF may be correlated with the cold gas inflows that are determined by the dark matter halos and environments.

Figure 20 shows the face-on and edge-on views of gaseous mass in the disk-dominated galaxies D2-D5 and bulge-dominated galaxies B2-B3 (marked by squares in Figure 16). They have similar stellar masses of $M_{\rm s}\simeq 10^{10.2}\ M_{\odot}$, but their sizes vary from $\sim 1.5$ kpc to nearly 10 kpc. There is a clear signature that the dramatic difference in their sizes originates from the difference in angular momentum of accreted gas. More compact galaxies are likely to be fuelled by gas inflows whose angular momentum is lower or removed sufficiently, thus generating a smaller gaseous disk (Figure 15), which is consistent with the theoretical expectation (e.g. Mo et al., 1998; Bullock et al., 2001). Moreover, the galactic wind driven by supernova and AGN feedback may also play a role, as suggested by Genel et al. (2015).

Because gas seems to be directly accreted into galaxy central regions in B2, it is then able to contribute directly to the growth of the bulge, shown in Figure 21. B2’s bulge keeps growing until $z=0$, thus the galaxy size changes little. It is plausible that low-mass bulge-dominated galaxies evolve along a similar compact pathway to massive ones, while more massive bulge-dominated galaxies evolve naturally either earlier or faster.

Two prototype halo-dominated galaxies H1-H2 (marked by squares in Figure 22) are discussed in this section. They evolve along extended and compact pathways that are shaped by mergers, especially major ones.

H1 (Figure 24) is a typical massive elliptical galaxy. At $z=1.5$, the progenitor of this galaxy has about $10^{10.6}\ M_{\odot}$ of stellar mass and a compact morphology of $r_{\rm e}\sim 0.7$ kpc, thus a typical nugget. It evolves in a similar way to bulge-dominated galaxies such that a disk is assembled via in-situ SF during $z=0.2-1.5$ before the major merger happens at $z\sim 0.2$. If no merger is involved in the evolution of this galaxy, it would become a galaxy of $M_{\rm s}\sim 10^{11.0}M_{\odot}$ and $r_{\rm e}\sim 4$ kpc, i.e., a massive bulge-dominated galaxy, according to its properties at $z=0.5$. The major merger transforms it to a halo-dominated elliptical galaxy with $M_{\rm s}=10^{11.3}M_{\odot}$ and $r_{\rm e}\simeq 10$ kpc. The growth of galaxy size is likely a natural outcome combining mergers and extended SF via gas accretion. The most dramatic increase of the galaxy size is driven by the final major merger, especially for galaxies along the compact pathway in Figure 22, generating a massive elliptical galaxy with a diffuse envelope. About $50\%$ of the stellar halo particles are from the satellite galaxy merging in and SF during the merger event; another $50\%$ come from the stars previously existing in the central galaxy. Although the bulge’s progenitor of such a galaxy is already quite massive, reaching $\sim 10^{10.5}\ M_{\odot}$, at $z=1.5$, it only contributes to a small fraction of the total stellar mass at $z=0$ because of the contribution from ex-situ stars accreted during the merger.

As shown in Figure 22, most of halo-dominated galaxies that evolve along the compact pathway become more massive by a factor of $\sim 2$ during $z<2$. Meanwhile, their sizes are $\sim 10$ times larger. Such an evolutionary pathway is consistent with “nuggets” that are believed to eventually become more extended elliptical galaxies. In contrast, the stellar masses of the galaxies following the extended pathway increase significantly by a factor of $\sim 10$, while their sizes become only a few times larger. Even for the most massive cases, there is a non-negligible fraction of halo-dominated galaxies formed from the extended pathway. This fraction increases significantly toward the low-mass end.

Figure 25 shows the evolution of a prototype halo-dominated galaxy that forms along the extended pathway. It is clear that an initially extended disk is destroyed by a major merger at $z\sim 0.7$, but a new disk with $32.4\%\ M_{\rm s}$ is generated in the later time, thus forming a Sombrero analogue. In such galaxies, a significant merger can disrupt the secular evolution at certain time, but it cannot shut it down completely. We therefore suggest that galaxies cannot be quenched directly by either mergers (e.g. Hopkins et al., 2006, 2008) or the growth of halos.

The compact-extended evolutionary pathway can be explained by the long-standing concept of the spin parameter (Fall & Efstathiou, 1980; Blumenthal et al., 1984; Mo et al., 1998; Dutton et al., 2007). The basic idea is that galaxy stellar disks form as a consequence of gas slowly cooling from a hot gaseous halo, in the meanwhile, maintaining its specific angular momentum. A remarkable scaling relation is found between half-mass radius of the galaxy’s stellar distribution and its virial radius, as well as the spin, of the galaxy parent halo out to high redshifts (Kravtsov, 2013; van der Wel et al., 2014; Shibuya et al., 2015; Somerville et al., 2018). Likewise, we also find a clear signature that galaxy sizes are partly controlled by halo angular momentum. Compact galaxies with a more massive bulge generally have a smaller spin than extended disk-dominated galaxies at fixed stellar mass, without being affected by mergers. This difference finally leads to different evolution along either a compact or extended evolutionary pathway in nature, as illustrated in Figure 26. Chaotic, violent instabilities, i.e., the so-called “compaction” phase (Dekel & Burkert, 2014), are likely to be involved in the compact evolutionary pathway, which may facilitate the growth of bulges. Moreover, there is a natural downsizing in the compact-extended evolutionary pathway: a compact phase occurs earlier in more massive galaxies. At $z=2$, this phase is already over for massive galaxies, forming massive compact galaxies, while it is still underway in low-mass ones even at $z=0$.

We suggest that less massive quiescent central galaxies are bulge-dominated galaxies that are common in ETGs in the mass range of $10^{10.5}M_{\odot}<M_{\rm s}<10^{11}M_{\odot}$. Similarly, Lagos et al. (2018) also reported a quiescent population in less massive galaxies that have not had any mergers in the EAGLE simulation. Therefore, not all compact galaxies (nuggets) evolve into elliptical galaxies. By breaking the degeneracy between bulges and halos, we showed that many massive compact galaxies become the bulges in bulge-dominated galaxies, as proposed in Dullo & Graham (2013); Graham (2013); Graham et al. (2015). A similar conclusion is reached in Wellons et al. (2015, 2016) using the Illustris simulation.

In the later phase, relatively dry mergers, start to be destructive for disk structures in galaxies; i.e., the nurture effect in Figure 26. In the general picture, compact galaxies are believed to eventually become more extended quiescent galaxies via the cumulative effect of minor mergers that drives the increase of massive quiescent galaxies via build-up of a diffuse envelope (Naab et al., 2009; Hopkins et al., 2010; Oser et al., 2010; Porter et al., 2014; Huang et al., 2016). This is partially due to the fact that the number density of quiescent galaxies increases by a factor of $\sim 10$ during $z<2$ in observations (Brammer et al., 2011), which cannot be sufficiently explained by the major merger rate during this time (Robaina et al., 2010; Brammer et al., 2011). Genel et al. (2018) have found a similar trend in the SF and quiescent galaxies from TNG100 that reaches a good agreement with observations (Shen et al., 2003; van der Wel et al., 2014). Mergers, however, are unable to account for the density evolution of less massive quiescent galaxies. Other primary processes, e.g., the formation of bulge-dominated galaxies suggested in Section 8.1, are required to quench the star-forming galaxies in the low-mass end to explain the remaining growth of quiescent galaxies since $z\sim 2$.

In our picture, classical bulges are compact structures formed mainly in the early phase, while slow rotator elliptical galaxies are diffuse objects that are dominated by halos formed in the later phase. Both the classical bulges and the cores of massive elliptical galaxies, however, are likely to be formed in similarly fast SF at high redshifts, which is evident by recent observations of red spiral galaxies (Hao et al., 2019; Guo et al., 2020b; Zhou et al., 2020). The difference between the bulge- and halo-dominated galaxies increases in their subsequent evolution, largely due to major mergers that lead to a sharp increase in both mass and size of halo-dominated galaxies.

Therefore, our results suggest that quiescent ETGs are composed of halo- and bulge-dominated galaxies. The bimodality in galaxy types is, thus, contributed by both nature and nurture processes. An accurate decomposition of bulges and halos is required to understand the formation and evolution of galaxies and their structures.