WorksPhysics and Astronomy
Unbound matter in the circum-halo medium: a common mediator of halo secondary bias
Abstract
Halo secondary bias describes the dependence of halo clustering on properties beyond mass. Here we report that a new environmental metric, the unbound matter fraction () in the – shell around a halo, can largely account for secondary bias. Using haloes with – at obtained from TNG300-1-Dark, we find that controlling for reduces the secondary bias of formation time, concentration, spin, shape and velocity anisotropy to a residual correlation of , outperforming other environmental metrics such as the tidal anisotropy for spin and velocity anisotropy. Applying a stacking method to the circum-halo medium (CHM), we identify two distinct flow patterns: low- haloes co-move with coherent flows, with the CHM converging onto the haloes; in contrast, high- haloes move through dynamically hot, partially virialized structures, with the CHM passing by rather than being accreted. This dichotomy points to two distinct regimes of secondary bias, a cold' regime and a hot' regime, in which tidal and inertial forces act as the main drivers, respectively, with the dominant mechanism migrating from the former to the latter as haloes assemble into massive structures. At fixed large-scale environment, further indicates the presence of nearby neighbours and the competitive accretion. Our results highlight the role of unbound matter as a mediator of environmental effects on haloes, and reveal the physical mechanisms that shape the CHM pattern through which these effects are mediated.
Introduction
In the -cold-dark matter (CDM) cosmology, dark matter haloes are the fundamental building blocks of cosmic structure and provide the sites where galaxies form and evolve (e.g. White & Rees [102]; Mo et al. [54]). Understanding the clustering of dark matter haloes is therefore central to both cosmology and galaxy formation. The dependence of halo bias on halo mass itself has a long theoretical history: it was first derived from the extended Press-Schechter formalism by Mo & White [52] and tested extensively against simulations, and was later refined by the ellipsoidal-collapse calculation of Sheth et al. (2001). Motivated by this success, most empirical descriptions of the galaxy-halo connection, including Halo Occupation Distribution (HOD) and Conditional Luminosity Function (CLF) models, assume that halo clustering is determined solely by halo mass (Berlind & Weinberg 2002; Yang et al. 2003). This assumption underpins many statistical approaches for interpreting galaxy surveys and connecting observed galaxies to the underlying dark matter distribution, making its validity central to modern models of the galaxy-halo connection (e.g. Wechsler & Tinker 2018).
Yet mass may not be the whole story: the ellipsoidal-collapse calculation already implied a secondary dependence of halo clustering on formation time through the local tidal field, as pointed out explicitly by Sheth & Tormen (2004), before such a dependence was discovered in simulations. This mass-only paradigm was eventually challenged by the discovery of halo assembly bias in N-body simulations, whereby haloes of the same mass exhibit different clustering amplitudes depending on their assembly histories (Gao et al. 2005). Subsequently, assembly bias has been identified for an increasingly diverse set of halo properties, including concentration, spin, shape, velocity anisotropy, subhalo occupation, and other structural and dynamical indicators (e.g. Wechsler et al. 2006; Gao & White 2007; Jing et al. 2007; Wang et al. 2007; Li et al. 2008; Faltenbacher & White 2010; Lazeyras et al. 2017; Mao et al. 2018; Xu & Zheng 2018; Sato-Polito et al. 2019; Han et al. 2019; Johnson et al. 2019; Chen et al. 2020; Wang et al. 2021; Balaguera-Antolínez et al. 2024; Ramakrishnan et al. 2025; Wang et al. 2026a,b}). Since this phenomenon is not solely caused by the halo assembly history, it is also broadly referred to as secondary bias. The growing list of secondary bias types suggests that halo clustering is sensitive to a much richer set of halo properties than halo mass alone.
Although the existence of assembly bias is firmly established, its physical origin remains an open question. Early studies mainly sought to explain individual forms of secondary bias through distinct physical mechanisms, such as peak theory (Zentner 2007; Dalal et al. 2008), splashback objects (e.g. Wang et al. 2009; Ludlow et al. 2009; Sunayama et al. 2016; Villarreal et al. 2017; Mansfield & Kravtsov
© 2026 The Authors 2020; Tucci et al. 2021), tidal suppression (e.g. Hahn et al. 2009; Wang et al. 2011; Shi et al. 2015; Salcedo et al. 2018; Mansfield & Kravtsov 2020; Smith et al. 2024), protohaloes (e.g. Lee & Moon 2024; Montero-Dorta et al. 2025; Nikakhtar et al. 2025), and cosmic web (e.g. Shen et al. 2006; Yang et al. 2017; Borzyszkowski et al. 2017; Musso et al. 2018; Montero-Dorta & Rodriguez 2024; Coloma-Nadal et al. 2024); see also Busch & White (2020) for a direct characterization of the nested hierarchy of density peaks. These mechanisms do not constitute a unified theoretical framework for halo assembly bias; rather, they imply that different forms of secondary bias may originate from distinct, yet closely related, physical processes operating between haloes and their surrounding large-scale environments (Salcedo et al. 2018; Mansfield & Kravtsov 2020).
Several studies have sought a unified framework rather than separate explanations for individual halo properties. An early step in this direction was taken by Wang et al. (2011), who proposed that the local tidal field regulates halo mass accretion and thereby simultaneously influences multiple secondary halo properties (see also Hahn et al. (2009)). In this framework, the correlation between halo internal properties and large-scale clustering stems from the strength of the local tidal field (Wang et al. 2007; Shi et al. 2015). Building upon this idea, Paranjape et al. (2018a) introduced tidal anisotropy as a more fundamental environmental variable (see also Ramakrishnan et al. 2019; Ramakrishnan & Paranjape 2020). They showed that the anisotropy of the surrounding tidal field accounts for much of the dependence of halo clustering on a broad range of secondary halo properties, suggesting that many manifestations of secondary bias are driven by the interaction between haloes and the anisotropic cosmic web (Borzyszkowski et al. 2017; Musso et al. 2018).
Although the tidal field and its anisotropy provide a successful environmental description of halo secondary bias, they are external quantities that characterize the large-scale environment surrounding a halo, and do not directly record how environmental forces modify halo properties. A promising clue was recently provided by Ma et al. [46], who showed that the spin bias of low-mass haloes changes qualitatively depending on whether unbound matter in haloes is included in the halo definition. Besides spin, unbound matter also affects the concentration bias, but has little effect on the shape bias (Ma & Li 2026). These findings suggest that unbound matter may provide the internal link missing from the purely external tidal description, i.e. a quantity that records how the environment acts on haloes. Whether unbound matter merely modifies individual bias measurements or traces a physical process that generates the secondary bias as a whole has not been addressed.
In this paper, we explore the role of the unbound mass fraction, , in halo secondary bias using the TNG300-1-Dark simulation. § 2 describes the simulation and the measurement of halo properties. § 3 shows that controlling for largely removes the secondary bias associated with the internal halo properties considered, and benchmarks against the two leading environmental indicators proposed in previous studies. § 4 identifies the circum-halo flow patterns traced by and proposes the physical mechanisms that shape them. § 5 summarizes our results and discusses implications for the galaxy–halo connection and observational tests.
Simulation and halo properties
The simulation and halo sample
Throughout this paper, we use the TNG300-1-Dark simulation, carried out as part of the IllustrisTNG project (Springel et al. 2018; Pillepich et al. 2019; Nelson et al. 2018, 2019; Marinacci et al. 2018; Naiman et al. 2018). The simulation evolves the density field using 2500 dark-matter particles, each with a mass of , in a periodic box with a side length of 205 . The simulation adopts the Planck-15 cosmology ([69]), with , , , with , and . Haloes are identified using the FoF algorithm with a scaled linking length of 0.2 (Davis et al. 1985), and subhaloes are identified using the Subfind algorithm (Springel et al. 2001). Subhalo merger trees are constructed using the Sublink algorithm (Rodriguez-Gomez et al. 2015). The centre of a halo is defined as the position of the most bound particle in the central subhalo identified by Subfind. The virial radius () of a halo is defined as the radius within which the enclosed matter density equals the critical density threshold given by the spherical collapse model (Bryan & Norman 1998), and the virial mass () of the halo is the total mass enclosed within . The virial velocity is defined as . Our sample consists of distinct haloes with –, identified at . The lower mass limit ensures that haloes are resolved with particles, sufficient for measuring the properties described below.
Secondary properties of haloes
We define the following internal properties of each halo, which characterize its assembly and structure, and have been found to correlate with halo clustering:
The half-mass formation redshift, (hereafter also referred to as the formation time), is the redshift at which a halo first reaches half of its final virial mass at . We determine this redshift by linearly interpolating the mass-redshift relation between two adjacent snapshots in which the halo mass crosses (Gao et al. 2005).
The concentration, , is computed from the virial velocity, , and the maximum circular velocity, , both measured using all particles, through the relation
where , is the radius at which is reached and is the scale radius. This concentration estimator is robust to variations in the detailed halo profile (Klypin et al. 2011).
The spin parameter, , is a dimensionless measure of the total angular momentum (Bullock et al. 2001), defined as
where is the magnitude of the total angular momentum of the particles within . Spin bias is sensitive to the definition of the spin parameter, particularly to whether unbound particles are included ([46]). We therefore calculate the spin parameter using two particle sets: bound particles () and all (bound plus unbound) particles () within . To determine whether a particle is bound to a halo, we use all particles within of its centre to compute each particle’s specific kinetic energy (including the Hubble flow) and specific potential energy (see [46] for details), with the zero point of the potential energy set at the most distant particle among those within . Tests with alternative choices (e.g. using an aperture of ) yield similar results. A particle is classified as unbound if its total specific energy (kinetic plus potential) is positive, and as bound otherwise. Note that particles beyond can be classified as gravitationally bound and may therefore be accreted into the halo in the future. We use the same classification when computing the halo properties defined below.
Shape parameters are obtained from the weighted inertia tensor,
computed iteratively ((Allgood et al. 2006)) using all bound particles enclosed within . Here, is the displacement of the -th particle from the halo centre along the -th spatial dimension in the simulation frame; the length of the -th ellipsoidal axis, , is proportional to the square root of the -th eigenvalue of , with the axes ordered such that ; the weighting uses the ellipsoidal distance , defined as
where is the displacement of the -th particle from the halo centre along the -th eigenvector of . Following common practice, we use the minor-to-major axis ratio, , to characterize the departure of the halo from sphericity, with small indicating an elongated shape and indicating a spherical shape.
The velocity anisotropy is defined as ((Binney & Tremaine 1987)), where and are the tangential and radial velocity dispersions, respectively. We calculate these dispersions by projecting the velocity of each particle along and perpendicular to the radial direction and then computing the standard deviation of each component over all bound particles within . A value of indicates an isotropic velocity distribution, while indicates radially biased orbits and indicates tangentially biased orbits.
We use the following properties to quantify halo clustering and characterize the surrounding environment, providing diagnostics of the origin of secondary bias:
The halo bias, , measured in a halocentric spherical shell spanning the radial range is defined as the ratio of the halo-matter cross-correlation function to the matter auto-correlation function within the shell:
where is the matter overdensity, estimated using the particles within the shell, and is the mean matter density of the Universe. In the main text, we focus on halo bias over the radial range 5–12 , a quasi-linear regime (e.g. (Gao et al. 2005)) that has been widely used in the literature to measure halo bias (e.g. (Chen et al. 2020; Zhang et al. 2025)). In Appendix C, we also examine the bias profile, , obtained using shells with finite thickness spanning .
The tidal anisotropy factor, , at a spatial location quantifies the asymmetry of the gravitational environment at that location ((Paranjape et al. 2018a)). We follow previous studies ((Hahn et al. 2007; Forero-Romero et al. 2014)) and compute from the deformation tensor, , defined as the Hessian of the gravitational potential (scaled by a factor of ):
where is obtained by solving the Poisson equation using the overdensity field derived from all particles in the simulation. Here, and are the Cartesian coordinates along the -th and -th dimensions, respectively, and is the gravitational constant. Since halos of different sizes are affected by tidal fields on different scales, we compute for each halo at a smoothing scale proportional to its virial radius. We smooth using a Gaussian window with a kernel size of , a choice found to yield the strongest signal of secondary bias ((Paranjape et al. 2018a)). We then use the smoothed overdensity field to compute and , evaluating the tensor at the halo centre to characterize the tidal environment of the halo. We diagonalize to obtain the eigenvalues and compute the tidal anisotropy factor as
where is the tidal shear, and is the overdensity at the location of the halo.
The environmental axis ratio, , is measured in the same way as the halo axis ratio (), but using all particles at halocentric distances in the range – ((Ma & Li 2026)). We omit the iterative procedure, setting for all particles in the computation of the inertia tensor (Eq. (3)) and evaluating the ellipsoidal axes directly from that tensor. Our tests show that including the iteration yields similar values of .
The unbound mass fraction
The unbound mass fraction, , in a spherical shell centred on a halo is defined as the total mass of unbound particles divided by the total mass of all particles within that shell. To illustrate the definition, Figure 1a presents two example halos with comparable

Figure 1. Two example halos with distinct unbound mass fractions. (a), distribution of dark matter in a slice of the simulation box with a thickness of . Here the grey background shows dark matter surface density in this slice. (b), (c), zoom-in views of the two example halos marked by black frames in panel (a). The two halos have comparable masses (), but contrasting unbound mass fractions of and 0.21, respectively, measured within . Each zoom-in panel is centred on the halo centre, with four dashed circles marking the halo-centric distances of 1, 2, 3, and , respectively; individual bound and unbound particles are represented by blue and red dots, respectively. This figure illustrates the contrast in unbound mass fraction between the two halos, and the different large-scale environments they reside in (see § 2.3 for details).
mass
fractions of unbound matter. (a), distribution of dark matter in a slice of the simulation box with a thickness of . Here the grey background shows dark matter surface density in this slice. (b), (c), zoom-in views of the two example halos marked by black frames in panel (a). The two halos have comparable masses (), but contrasting unbound mass fractions of and 0.21, respectively, measured within . Each zoom-in panel is centred on the halo centre, with four dashed circles marking the halo-centric distances of 1, 2, 3, and , respectively; individual bound and unbound particles are represented by blue and red dots, respectively. This figure illustrates the contrast in unbound mass fraction between the two halos, and the different large-scale environments they reside in (see § 2.3 for details). masses () but contrasting (measured within 4 ). Panel (a) shows their large-scale environments: the low- halo resides in a void, whereas the high- halo lies near a node, at the root of a filament. Panels (b) and (c) zoom in on the high- and low- haloes, respectively: both haloes have a significant number of bound particles within ; around the high- halo, the projected matter density is notably higher and dominated by unbound particles, whereas the low- halo is embedded in a sparser field consisting mainly of bound particles. The contrast between the bound-dominated interior and the unbound-rich outskirts motivates examining in radial shells. To examine the radial dependence, we calculate for each halo in shells spanning four radial ranges: (0–1) , (1–2) , (2–3) and (3–4) . Figure 2 shows the median as a function of halo mass. The unbound mass fraction is higher for lower-mass haloes and increases with distance from the halo centre. Within the virial radius, is less than 0.008, consistent with results in the literature (see fig. 3 in Behroozi et al. 2013). In the outermost shell, (3–4) , reaches 0.8 for low-mass haloes but only about 0.3 for high-mass haloes, a mass dependence reflecting the stronger gravity of more massive haloes in competing against external forces, as discussed in § 4.
Figure 2. Fraction of unbound mass in and around haloes. Here we show the unbound mass fraction () as a function of halo mass () measured within shells spanning different halo-centric distances: 0–1 (blue), 1–2 (green), 2–3 (orange) and 3–4 (red). Solid lines show the median values and the shaded regions the 25th–75th percentiles. This figure shows that is higher for lower-mass haloes and increases with distance from the halo centre (see § 2.3 for details).
As we will see, encodes most of the secondary bias, while other assembly and structural properties produce little additional bias once is taken into account. This identifies as a mediator through which environmental effects propagate into haloes, and thereby as a route to the origin of secondary bias.
Secondary bias of haloes
as a mediator of secondary bias
To see the relation between and secondary bias of haloes, we partition the simulated haloes into bins of , and quantify the relation between and halo bias within each mass bin. We divide the haloes in each bin into quartiles based on their values, and we compute the average bias, (Eq. (5)), among haloes for each quartile. Figure 3 shows average in each quartile as a function of (coloured curves), adopting four different radial ranges in the measurement of (panels a–d). For reference, the average over all haloes, without binning , is shown by a black curve; a random subset of individual haloes is shown as scatter points, with their colour-coded.
Figure 3. Relation between unbound mass fraction and halo bias . (a)–(d) show the results obtained using measured in shells spanning different halo-centric distances. In each panel, individual dots in the background represent randomly selected haloes, coloured by the value measured in the corresponding shell, according to the colour bar on the top. Curves show the average halo bias of haloes in bins of , separately for haloes in the four quartiles of in each mass bin (coloured solid) and for all haloes in that bin (black dashed). This figure shows that haloes with higher are more strongly clustered, and that such a bias– correlation persists in all radial ranges and halo masses examined (see § 3.1 for details).
As expected, the primary bias, namely the relation between and , shows a positive slope with a flattening at low masses (). At a fixed , haloes with a larger are more strongly clustered, with low-mass haloes in the top quartile of having a bias comparable to the mass-only bias of haloes with . Such a tendency for higher- haloes to exhibit higher bias is consistent with the difference between the environments in which the two example haloes reside (Figure 1b,Figure 1c). Notably, the – relations for different quartiles are clearly separated over the entire range of examined here, and the separation remains pronounced in all radial ranges. The most pronounced – correlation is observed for measured within 2–3 (Figure 3c). In comparison, assembly bias measured with respect to other internal properties of haloes (e.g. those listed in § 2.2) is generally weaker and/or mass-dependent (see e.g. Gao et al. 2005; see also the correlation analyses below).
The strong correlation between and halo bias suggests an origin of the secondary bias in which the environment shapes , and through it, the internal (structural and assembly) properties of haloes. If indeed serves as such a mediator between environment and haloes, this immediately yields a testable prediction: the correlation between bias and any internal property of haloes should vanish, once is fixed. A common approach to quantify such a dependence graph is through the conditional correlation coefficient, , defined as (e.g. Ramakrishnan et al. 2019):
for variables and conditioned on the mediating variable , where , and are the (unconditional) pairwise Spearman correlation coefficients among the variables. A value of close to zero indicates that the correlation between and is largely accounted for by their correlation with , implying conditional independence between and given . In our application, is the unbound fraction (), is the bias factor (), and is one of the internal properties of the halo (e.g. , , , or ; see § 2.2 for definitions). Unless explicitly stated, hereafter, we use in the radial range of 2–3 as our default choice, where the – correlation is strongest (Figure 3c; see also Appendix § 6 for alternative apertures). The uncertainty for the conditional correlation coefficient is estimated as the standard deviation among 50 bootstrap resamples with replacement.
Figure 4a shows , the correlation between each internal property and halo bias , which quantifies the secondary bias that has been discovered in previous studies. Older haloes (with higher ) are clustered more strongly (exhibiting higher ) at the low-mass end, but this trend disappears at the high-mass end (e.g. Gao et al. 2005). behaves similarly to at the low-mass end, where the halo profile is found to be closely tied to the formation history (Wechsler et al. 2002; Ludlow et al. 2013), but the sign of the correlation reverses at the high-mass end (e.g., Gao & White 2007), a consequence likely arising from the configuration of the protohalo determined by the distribution of progenitor haloes beyond that in the main branch (Mao et al. 2018; Wang et al. 2020, 2024), leading to a delayed response for the formation time (see discussion in Appendix § 9). Haloes with lower are more strongly clustered at the low-mass end, a trend that reverses at (e.g., Sato-Polito et al. 2019). This inversion of is insensitive to the set of particles in use: additionally including unbound particles produces a correlation with a positive sign over the entire mass range (Ma et al. [46]), discussed further in Appendix § 9. Rounder haloes, and haloes with more tangentially biased orbits, are clustered more strongly over the entire mass range (e.g., Faltenbacher & White 2010). For comparison, we show , the correlation between the unbound fraction and halo bias , which maintains a positive sign and a stable magnitude of – across the entire mass range examined here. Importantly, is much larger than all the other , reinforcing the significance of the unbound fraction in tracing secondary bias. The significance of , together with the constancy of its sign and magnitude across halo mass, yields a clear expectation: if indeed mediates the effect of the environment on an internal property of a halo, , then
Figure 4. Spearman correlation coefficients between halo bias, and halo internal properties. (a), correlations between halo bias and each internal property : formation time (, purple), concentration (, red), spins in two definitions (, orange; , light-blue), shape parameter (, green) and velocity anisotropy (, blue). For comparison, the correlation between and is also shown (black). (b) Correlations between and the internal properties. (c) Correlations between and the internal properties after controlling for . In each panel, shaded region around each curve shows the bootstrap error; the horizontal dotted line marks zero correlation. This figure shows that is the strongest predictor of halo bias, and that controlling for largely removes the residual biases of internal halo properties (see § 3.1 for details).
(i) and should exhibit a significant correlation, for any that exhibits significant secondary bias;
(ii) the sign of , and the mass dependence of its sign and magnitude, should be similar to those of ;
(iii) the conditional correlation, , should vanish.
Fig. 4(b) and (c) show and for various internal properties of haloes, confirming the expectations listed above: all the internal properties exhibit a significant correlation with ; the sign of is the same as that of , and for , and , the sign of is inverted at nearly the same halo mass as for ; the magnitude of also closely follows that of , albeit lower by a factor of –; all the conditional correlations, , are close to zero (with magnitude ) over the entire mass range examined here. Notably, the correlation between velocity anisotropy () and almost completely disappears after conditioning on , suggesting that outperforms other environmental indicators, such as the tidal anisotropy ((Ramakrishnan et al. 2019)), in determining the velocity structure of haloes (see § 3.3 and Appendix B for more discussion). The most obvious residual, at the level of , is seen only for , and . These three internal properties are more susceptible to peculiar events, such as ‘splashbacks’ (e.g. (Li et al. 2013)), that take place long ago in the histories of the haloes, and are thus less completely tracked by the mediator, , which is measured at (see § 3.4 for quantitative discussion).
To reassure the reader that these conclusions are not an artifact of the limited ability of correlation analysis to capture non-linear relations, we also adopt a conventional metric of secondary bias. We take the top and bottom quartiles of the distribution of each internal property in each bin of , compute the average for haloes in each quartile, and use the difference of the average between the two quartiles as the metric of (unconditional) secondary bias. To obtain the secondary bias conditioned on we assign a weight to each halo in a quartile so that the weighted distribution of for haloes in that quartile is identical to that of all haloes within the same bin of (see e.g. (Zhang et al. 2025), for a similar weighting scheme); the average of using these weights is then computed for haloes in each quartile, and the difference between the two quartiles measures the secondary bias conditioned on .
Figure 5 shows the unconditional (dashed curves) and conditional (solid curves) secondary biases. As expected, all six internal properties show significant unconditional secondary bias; the conditional ones, however, nearly vanish for , , and , and are largely reduced for the other properties as well. These results support the role of as a mediator in propagating the environmental influence on internal properties of haloes.
Figure 5. Halo secondary bias with or without controlling for . (a)–(f), average halo bias in bins of for haloes in the top and bottom quartiles of each internal property, , as indicated in the bottom-right corner of each panel. In each panel, coloured dashed curves show the bias directly averaged from the haloes in the quartiles of , while coloured solid curves show the bias averaged with weights that align the distributions of each quartile to that of all haloes in the corresponding bin. Shaded regions or error bars associated with each curve show the bootstrap errors. For reference, the black dashed curve shows the halo bias averaged over all haloes. This figure confirms the result from correlation analyses (Figure 4) that controlling for largely removes the secondary bias associated with internal halo properties (see § 3.1 for details).
Physically, the correlation between and internal properties of haloes also implies a causal direction, since the large-scale environment is not likely to respond directly to the evolution of a halo. Thus, our correlation analysis, together with the physically implied causal direction, points to a picture in which a denser environment induces a higher unbound fraction around haloes, and such an increase in the unbound fraction subsequently influences the assembly and structure of haloes. In what follows, we compare our findings with those obtained using alternative environmental metrics. This places our findings in a broader context, and helps clarify how the large-scale environment affects haloes through .
Linking fun to alternative environmental metrics
Previous studies have proposed explanations for secondary bias by linking the internal properties of haloes to alternative indicators of environment. These studies identify a key factor, the tidal field, that can alter the assembly and structure of haloes and thereby generate the secondary bias. Descriptors for the tidal field include direct measurements of the tidal tensor (e.g. (Hahn et al. 2007; Forero-Romero et al. 2014)), the metrics derived from the tidal tensor, such as its eigenvalues ((Wang et al. 2011)) and anisotropy (, see § 2.2), and indirect indicators such as whether or not a halo has experienced a splashback event in the past (e.g. Wang et al. 2009; Li et al. 2013; Sunayama et al. 2016; Mansfield & Kravtsov 2020; Tucci et al. 2021). As we will show in the following, these indicators are not only linked to , providing complementary viewpoints on how environment affects haloes, but also account for the residual of secondary bias that is missed by , particularly that involving the three internal properties (, , ) whose residual is small but remains noticeable after controlling for (at a level of ; see Figure 4c).
Tidal anisotropy
As shown by Paranjape et al. (2018a) and Paranjape (2021), the tidal anisotropy (), which quantifies the asymmetry in the gravitational environment of a halo, can effectively distinguish between different cosmic-web environments the halo resides in: an isotropic environment, such as voids and nodes, yields a small , while an anisotropic environment, in particular that around massive filaments and sheets, yields a large ; a value of and can indicate highly isotropic and anisotropic regions, respectively (e.g. Paranjape et al. 2018a). Most relevant to this study, has been proposed as the driving factor that determines the assembly and structure of haloes at a given mass, while the secondary bias, namely the correlation between the large-scale bias factor () and internal properties of haloes, only emerges due to the correlation between and (Paranjape et al. 2018a; Ramakrishnan et al. 2019). This is very similar to the role of in controlling the secondary bias (§ 3.1), indicating that higher tends to be seen in more anisotropic environments, and vice versa. In Figure 6a, we show the distribution of individual haloes in the – plane, with colour indicating the tidal anisotropy . Massive haloes () exhibit low values, consistent with their preference to reside within node environments (e.g. Paranjape et al. 2018a). In contrast, lower-mass haloes exhibit systematically higher values, as well as much wider scatter at fixed mass, reflecting the diverse environments they inhabit, as already evidenced by the two examples shown in Figure 1(b) and (c). Haloes with higher tend to be found in more anisotropic environments (showing higher values), and such a correlation is observed across the range of examined here. Lower-mass haloes appear to exhibit stronger correlations between and , and in particular, haloes with in their neighbouring regions of 2–3 often show . Figure 6(b) provides a quantitative confirmation: a positive correlation coefficient is present across the entire mass range. Adopting measured within 2–3 , is constantly high across the entire mass range, with haloes at the low-mass (high-mass) end showing slightly stronger (weaker) correlations of (0.6). While the overall trends remain consistent across different radial ranges, measured within 2–3 shows the strongest correlation with in nearly all mass bins, providing further justification for using measured within 2–3 as our default indicator of halo environment.
Figure 6. The relation between unbound mass fraction and tidal anisotropy (). (a), (measured within 2–3 ) versus halo mass for 10,000 randomly selected individual haloes (plus markers), coloured by . The solid curve shows the median of haloes in bins of , and the shaded region shows the 25th–75th percentiles. (b), Spearman correlation coefficients between and in bins of halo mass, with measured within different radial ranges (coloured curves). This figure shows that is strongly correlated with , indicating a role of as a mediator of the environmental effects on haloes (see § 3.3 for details).
The constantly high – correlation (adopting measured within 2–3 ), together with the role of in controlling the secondary bias, immediately suggests two implications: (i) the pattern of the correlation between and an internal property of a halo mirrors that between and this internal property (i.e. and should share the same sign and mass dependence); (ii) conditioning on either or should largely account for the secondary bias, while conditioning on both should provide little incremental suppression of the secondary bias. In Appendix B and Figure B1, we examine the correlation between and halo secondary properties, consistent with expectation (i). To examine the expectation (ii), we consider the chain of correlation coefficients with incremental conditioning: , the unconditional correlation; and , the correlation conditioned on and , respectively; , the correlation conditioned on both and (see Kendall & Gibbons 1990):
Figure B1. Spearman correlation coefficients between tidal anisotropy and halo properties. Each coloured curve shows the correlation between and an internal halo property as a function of halo mass. For reference, the correlation between and bias is shown as the black curve.
In Figure 7, we show the correlation coefficients with incremental conditioning in bins of . Compared with the unconditional correlations, conditioning on either or substantially reduces secondary bias. The performance of conditioning depends on the specific halo property under consideration: alone almost totally eliminates the secondary bias in spins ( and ) and velocity anisotropy (), as already demonstrated in Figure 4 and 5; , however, outperforms in controlling the secondary bias in formation time (), concentration (), and axis ratio (; at ); conditioning on both provides additional reduction in the residual of secondary bias, but the improvement over conditioning on alone is modest. We caution, however, that the exact value of a correlation coefficient involving depends on the definitions adopted for the properties being examined (see Appendix B for details).
Figure 7. Spearman correlation coefficients between bias and internal properties of haloes with different conditioning variables. Each panel shows the correlation between and one of the six internal properties of haloes. Results using four combinations of conditioning variables are shown: no conditioning (grey diamonds), conditioned on either (orange dashed), or (green dashed), or both (black solid). For clarity, we show only the bootstrap errors for the last combination using error bars. The horizontal grey dotted line indicates zero correlation. This figure shows that conditioning on either or removes most of the secondary bias, while conditioning on both leads to little improvement, supporting the role of as a mediator of the effects of environment characterized by onto haloes (see § 3.1 for details).
The constantly high correlation between and , and their consistent performance in controlling secondary bias, suggest a causal direction, : a tidal field with larger anisotropy leads to stronger shear motion between the halo and its circum-halo medium, leaving a smaller fraction of the mass to be bound to, and accreted by the halo. In this scenario, tidal anisotropy is the cause, unbound matter is the mediator, and the variation in the assembly and structure of haloes is the outcome. In §4, we will examine the underlying processes traced by and show that such a causal relationship holds, and only holds, for a subset of haloes that ‘co-move’ with their circum-halo medium.
Splashback events
A splashback event occurs when a halo, after an excursion as a satellite subhalo of another halo, is later ejected (e.g. [25] (Ludlow et al. 2009)). The population of splashback haloes is found to be partly responsible for the secondary bias, particularly for those with low (e.g. (Wang et al. 2009; Li et al. 2013; Sunayama et al. 2016; Mansfield & Kravtsov 2020; Tucci et al. 2021)). As with , it is therefore expected that also correlates with the occurrence of splashback events in the histories of haloes, and such a correlation may partly explain the secondary bias introduced by the splashback population. To quantify this, we identify all splashback events for each halo in our sample by tracing back in time along the main progenitor branch of the merger tree rooted in the central subhalo of the halo in question, and we define the splashback redshift, , of the halo as the last redshift at which the halo was a satellite subhalo. Since a splashback event at too early a redshift may not be relevant for the current state of the halo, we limit the range of by defining haloes with as splashback haloes; the remaining haloes are classified as non-splashback. This criterion results in a fraction of of haloes at the low-mass end being classified as splashback haloes. Figure 8 shows as a function of , separately for the splashback and non-splashback populations. At , the difference between the two populations in is negligible, while at
Figure 8. versus halo mass for splashback and non-splashback haloes. Solid curves show the medians of , with shaded regions indicating the 25th–75th percentiles, for splashback (blue), non-splashback (red) and all haloes (black) in bins of halo mass. This figure shows that splashback haloes have systematically higher (see § 3.4 for details).
, the splashback haloes possess about twice the of non-splashback ones. Such a difference may be expected: a lower-mass halo is more likely to be subdominant in gravity among the structures in its local environment, making its loosely bound exterior more susceptible to being stripped away once it makes an excursion around or into another structure.
To measure the correlations with the effects of splashback controlled, we repeat the correlation analyses for Figure 4 separately for the splashback and non-splashback populations, and we show the resulting correlation coefficients in Figure 9. Consistent with that reported by Tucci et al. (2021), excluding splashback haloes removes the - correlation at the low-mass end (). The correlation coefficients of the splashback population are larger in amplitude than those of the non-splashback population for properties , , and across the entire mass range, for at , but not for and (Figure 9a, Figure 9d), indicating that splashback accounts for a portion, but not all, of the secondary bias. The correlation coefficient (Figure 9b, Figure 9e) gives some clues about why the effect of splashback on secondary bias depends on the halo property in question: the correlation coefficients for , , and are much higher for the splashback population than for the non-splashback population; in contrast, is not enhanced, and in fact shrinks for the splashback population.
Figure 9. Comparison of the Spearman correlation coefficients for splashback and non-splashback haloes. (a)–(c), unconditional and conditional Spearman correlation coefficients similar to Figure 4, but for splashback haloes only. (d)–(f), similar but for non-splashback haloes only. In each panel, the horizontal dotted line indicates zero correlation. This figure shows that controlling reduces secondary bias for both splashback and non-splashback haloes, although the residual bias is more pronounced for the splashback haloes (see § 3.4 for details).
As shown in Figure 9(c) and (Figure 9f), the correlation coefficients after controlling for are largely reduced for both the splashback and non-splashback populations. As with the tidal anisotropy , this indicates a causal direction, splashback : splashback events alter the fractions of unbound matter surrounding the haloes, and this matter in turn affects the assembly and structure of these haloes. The mediator role of is again implied in this causal chain, which, together with the chain inferred in § 3.3, suggests that is a sensitive tracer of physical processes operating at different spatial and temporal scales around haloes. In the following, we examine these processes in more detail.
Physical processes traced by the unbound matter
The above correlation analyses imply a mediatory role of in linking the environments and haloes, but have not yet revealed the underlying physical mechanisms through which the unbound matter operates on haloes. These analyses, however, do provide directions for searching for the relevant physical processes. At scales of , anisotropic tides, as characterized by at these scales, accumulate sheared motion, stretch the matter surrounding haloes, reduce its boundedness, and alter the accretion flow (§4.1); at smaller scales, interactions between haloes, with mergers and splashback events as the extreme cases, have a more direct and more diverse impact, injecting substructures into haloes, limiting the accretion onto haloes, or stripping material from haloes (§ 4.3). These processes point to a common scenario, in which the environment modifies the flow pattern of the circum-halo medium (CHM), which in turn affects the accretion rate and geometry, and hence the internal properties of haloes. Matter accelerated to a high velocity relative to the motion of a halo is unbound from the halo by definition, and is thus captured by .
tracing the divergence and vorticity of circum-halo flow
The pattern of circum-halo flow
To gain an intuition for how the flow pattern of the CHM is related to , in Figure 10, we overlay the streamlines of peculiar velocity on the density field (grey scales) of the slice containing the same two example haloes (Figure 1). The two haloes, whose unbound fractions are extreme ( and 0.87 for the low- and high- cases, respectively; measured within ) are shown zoomed in to highlight their local environments (Figure 10b, c), with the coordinates and the velocity field measured relative to the position and peculiar velocity of the corresponding halo (indicated by an arrow in Figure 10a), and with the Hubble expansion included, consistent with the definition of binding energy (§ 2.2). For clarity, we hereafter refer to the coordinate system of the simulation and to that of the zoom-in views as the simulation frame and the halo-centric frame, respectively, and denote the coordinates (velocities) in the two frames by () and (), respectively. All velocities, both those encoded in the streamlines and those shown as arrows for the haloes, are colour-coded with the common colour bar of Figure 10.

Figure 10. Velocity fields around the two example haloes. Here we show the projected velocity field using streamlines in the same slice (panel a), and around the same two example haloes with extreme values (panels b, c), as those in Figure 1. The surface density field (grey background in panel a) is the same as that in Figure 1(a). Velocity in panel (a) is the peculiar velocity obtained from particles in the simulation frame, and in panels (b) and (c) is physical velocity (peculiar plus Hubble flow) relative to the example halo. The peculiar velocities of the example haloes are marked by arrows in panel (a). Both streamlines and arrows are colour-coded by the magnitude of the projected velocity, according to the colour bar on the top. This figure reveals the distinct flow patterns around haloes with different values (see §4.1 for details).
Comparing the two cases in the two frames reveals striking differences. (i) In the simulation frame (panel a), the low- halo is embedded in a coherent, large-scale (), high-speed () bulk flow whose velocity is comparable to that of the halo. The halo is therefore a ‘co-moving observer’ of the ambient medium, so that the relative velocity between the halo and the CHM is small. This small relative velocity implies that the CHM is effectively bound to the halo, which results in a low in the CHM and in a halo-centric flow pattern converging on the halo (panel c). (ii) In contrast, the high- halo is embedded in a convergent, low-speed (), filamentary structure in the simulation frame (panel a), while the halo itself moves much faster than the flow (). The halo therefore behaves as a ‘subhalo-like observer’ within a partially virialized, dynamically hot structure, mimicking the condition that would be experienced by a subhalo falling into a well-virialized host halo. The motion of the CHM relative to such a halo is set by the virial velocity of the massive host structure, yielding a relative velocity of large amplitude. Consequently, only a small fraction of the CHM can be bound to the halo, the accretion onto the halo is suppressed, and the halo-centric frame shows a unidirectional, non-convergent flow (panel b).
These two extreme cases show that the flow pattern of the CHM is traced by : a low- halo tends to be embedded in a coherent, dynamically cold flow in the simulation frame and to co-move with it; in contrast, a high- halo tends to move randomly within a partially virialized, dynamically hot structure, so that it behaves like a subhalo whose motion is largely randomized with respect to that structure. In the halo-centric frame, the flow pattern of the CHM therefore has a distinct divergence (i.e. ), converging onto the low- halo and divergently passing by the high- one.
To test whether the distinct flow pattern around these cases is representative of the halo population in general, we take haloes from each bin of and select subsamples of haloes whose values are in the upper and lower quartiles within the same bin. We then stack the halo-centric velocity fields around each subsample, using the stacking method developed by Ma et al. [46]. Because we also wish to examine the spin of the flow (see below), the alignment scheme is chosen so that the spin vectors of haloes (computed from the same set of particles as used for ) are aligned, while the axis is chosen to be aligned with the direction of the halo motion. In Figure 11, we show the stacked velocity field in a slice perpendicular to the aligned spin vector for each subsample, with the velocities projected onto the slice. To aid the discussion, we also obtain the halo-centric profiles of the radial velocity (, with a positive sign indicating outflow) and of the tangential velocity (, perpendicular to and to the spin vector) by stacking the corresponding velocity component of the CHM in spherical shells centred on the haloes in each subsample (see Ma et al. [46]). The profiles so obtained for subsamples with different and are shown in Figure 12. The radial and tangential velocity profiles provide complementary diagnostics of convergence and rotation.
Figure 11. Stacked velocity fields around haloes. (a)–(c), stacked velocity fields around haloes in the upper quartile of in each bin of . (d)–(f), similar but for haloes in the lower quartile of . Here the field around each halo is the physical velocity (peculiar plus Hubble flow) in the rest frame of the halo (referred to as the halo-centric frame), measured in a slice of thickness perpendicular to the page (the – plane), and projected onto the page. The stacking follows the method developed by Ma et al. [46], with the halo-centric frame aligned so that their spin vectors, computed from all particles within , point out of the page, and the velocity vectors of the haloes point in the direction. In each panel, the arrows of the streamlines show the direction and the colour encodes the magnitude of stacked velocity (normalized by the average virial velocity of haloes being stacked); the two black arcs indicate the halo-centric distances of and , respectively. This figure shows that haloes with different exhibit flow patterns with distinct divergence and vorticity in their circum-halo medium (CHM), indicating that environment regulates halo accretion through modifying the flow pattern of the CHM (see §4.1 for details).
Figure 12. Stacked radial and tangential velocity profiles of haloes. Here we show the physical velocity in the rest frame of the halo, stacked in spherical shells at different halo-centric distances around haloes in each bin of and . Two components of the velocity, the radial component (panels a–c; projected onto the direction from halo center to the particle, with a positive value indicating outflow) and the tangential component (panels d–f; projected onto the plane perpendicular to both and the spin axis of the halo), are shown separately. The figure is obtained in a similar way to Figure 11, using the stacking method of Ma et al. [46] and the same alignment strategy. In each panel, the vertical dashed line marks the average virial radius of haloes being stacked, while the vertical shaded region indicates the radial range of – where is measured. This figure quantifies the flow patterns of CHM revealed in Figure 11, with and serving as proxies for the (integrated) divergence and vorticity of the flow, respectively (see §4.1 for details).
The stacked fields are consistent with the pattern found around these two example haloes (Figure 10b, c), and reveal more:
(i) The high- subsamples (Figure 11a–c) all exhibit a passing-by flow out to , but a clear mass dependence is visible: the more massive haloes (panel c) have a lower CHM velocity (in the halo-centric frame, scaled by ), and the velocity direction is more radial within –. The physical interpretation of this mass dependence is that a more massive halo is more likely to be the dominant source of gravity, so that the random motion induced by the surrounding structure is less significant than that induced by the halo itself. The lower scaled velocity of the CHM around more massive haloes also implies that the CHM is more likely to be bound to the halo, which explains the mass dependence of (Figure 2). The radial profiles provide complementary evidence: the highest- quartile has outward median motion outside in the two lower-mass bins (Figure 12a, b), but retains an infall dip at high mass (panel c).
(ii) The low- subsamples (panels d–f) all exhibit a flow converging onto the halo centres, but the range over which inflow dominates is mass dependent. A circle-like ridge at which the radial velocity reverses sign (the turnaround radius; see Mo et al. [54]) is present around each subsample, and the radius of the ridge, , expands from to as increases from to . More massive haloes have larger inflow speeds within , and the inflow is more radially biased, consistent with the fact that CDM haloes exhibit a universal pattern of two-phase assembly and that more massive haloes transition to the slow-accretion phase later (e.g. Zhao et al. 2003; Mo et al. 2024). This mass dependence of the inflow is also seen in the radial velocity profile of the lowest- quartile: a negative- minimum is reached in the – shell, with a deeper minimum at higher mass (Figure 12a–Figure 12c). At low and intermediate mass, a clear infall dip outside is confined to this lowest quartile.
(iii) The CHM exhibits spinning motion in both the low- and high- subsamples, but the effect on the halo depends on whether that material can be retained. At low mass, rotation in the high- stack is superposed on a stronger passing-by flow (Figure 11a, Figure 11d). The tangential profiles resolve the radial dependence of this contrast (Figure 12d): the highest- quartile has larger outside , indicating a higher vorticity () of the CHM, yet a lower inner peak within than the lowest quartile. This reversal is consistent with high-angular-momentum material passing through the CHM without being retained by the halo. The low- flow carries less tangential motion in the CHM, but its inward motion (i.e. more negative divergence) allows angular momentum to be delivered to the halo. This interpretation connects the outer flow to the opposite correlations of with and
at low mass (Figure 4b), and hence to the inverted low-mass bias of (Figure 4a). At intermediate and high mass, the high- population no longer has a lower inner tangential velocity than the lowest quartile, indicating that angular momentum is effectively delivered to the halo along with the enhanced accretion at these masses.
Two regimes of secondary bias
The distinct flow patterns around the two populations with low and high , and the preference of the populations to reside in distinct large-scale environments (indicated by the correlation of with and ; see § 3), lead to a hypothesis that two regimes of secondary bias exist. In this hypothesis, the large scale environment (as characterized by and ) determines the driving factors that regulate the flow pattern (including the divergence and vorticity of CHM), and the flow pattern in turn determines the accretion onto haloes and hence their internal properties.
To illustrate the two regimes, and to demonstrate they are physically distinct, we seek the forces that drive the CHM motion. We refer the reader to Figure 13 for a schematic outline. Consider a mass element in the CHM surrounding, or about to pass by a halo in question. In the halo-centric frame (i.e. the halo’s rest frame, centred on the halo), this mass element is subject to a net force of
Figure 13. A schematic illustration of the two regimes of secondary bias. The two regimes, the ‘cold’ (numbered a; corresponding to low and small halo bias) and ‘hot’ (numbered b; corresponding to high and large halo bias) regimes, are characterized by the distinct dynamical hotness of the large-scale environment in which the target halo resides, producing different force cancellation, leaving distinct forces that regulate the flow pattern of the circum-halo medium (CHM) and the accretion onto the halo. Secondary bias in the cold and hot regimes is thus driven by different mechanisms, and a halo evolved from the former to the latter regime may have experienced a switch in the dominant mechanism. Notations: , the uniform part of the gravitational force; , the tidal (i.e. differential) part of the gravitational force; , the inertial force in the halo’s rest frame, produced by the acceleration () of the halo; , the pressure force from the random motion in a partially virialized structure. See §4.1 for details.
where all force components are measured per unit mass. Here, is the gravitational force averaged over all mass elements in the CHM (i.e. the uniform part of gravity across the CHM); is the tidal force, describing the differential gravity across the CHM of finite size; both and include the contribution from large-scale environment (i.e. the matter outside the CHM, plus the cosmological constant), and the halo itself; is the ‘pressure’ force caused by random motion (‘thermal pressure’) when the CHM belongs to a structure with a certain degree of virialization; and is the inertial force, which is a virtual force defined to be the negative of the halo acceleration. In the special case that a halo moves in an orbit around some centre, reduces to the centrifugal force; in general, the halo may not have a well-defined orbit (e.g. Figure 13a–Figure 13b).
We consider the following two extreme cases, as identified through their distinct , CHM flow pattern, and large-scale environment (§ 4.1.1), and examine each term that contributes to :
The ‘cold’ regime, where low- haloes co-move with their CHM embedded in coherent flows at large scale (e.g. voids; Figure 13a). In this case, the condition of coherent motion implies cold dynamics (i.e. the bulk motion dominates over the random motion, in terms of the kinetic energy partition; see e.g. Ma et al. [46] for quantifiers), so that the ‘pressure’ force is negligible,
In addition, the ‘co-moving’ motion implies that the halo and the CHM share the same acceleration of the coherent bulk flow, so that the uniform part of the gravitational field is canceled by the inertial force in the halo-centric frame,
The vanishing or cancellation of these forces leaves the tidal (differential) gravity the only force that can drive the relative motion. In this regime, whenever the halo’s own gravity dominates over the external tide, the CHM is naturally gathered onto the halo, giving rise to the infall-dominated pattern (Figure 11d–Figure 11f). The tidal field is therefore identified as the dominant factor regulating the motion of CHM relative to the halo, and plausibly underlies the secondary bias of low- haloes.
The ‘hot’ regime, where high- haloes move like subhaloes hosted within quasi-virialized, massive structures (high , e.g. massive filaments or outskirts of nodes). In this regime, the CHM is part of the virialised structure, while the CHM originally associated with the halo has already been stripped away (Figure 13b). Hence, the virialization condition implies hot dynamics in the host structure, and that the pressure support due to the random motion cancels the external gravity on the CHM from the large-scale environment:
leaving only the inertial force to operate on the CHM, and to compete with the gravity of the halo itself when the CHM is passing by the halo. Assuming the host structure has a mass , the inertial force generated on the CHM by the subhalo-like orbital motion of the halo in question is
where is the distance of the halo to the centre of the host structure, and is the gravitational constant. The external tidal force on the CHM of a finite size at a distance from the host structure is
where is the largest eigenvalue of the tidal tensor (see Eq. (6)). For CHM that is passing by, or is about to pass by, the halo, , so that the ratio of their magnitudes is
where a few times by the ‘circum-halo’ definition. Thus, the subdominance of the external tidal force, , holds for any halo whose size is much smaller than its distance to the centre of the host structure. Thus, the passing-by motion of the CHM is mainly driven by the inertial force, not the tidal force. Although
we show only the difference in the instantaneous forces here, its cumulative effect over time gives rise to an actual relative velocity between halo and CHM, as seen in Figure 11a–Figure 11c. Consequently, the inertial force that arises from the assembly of massive, dynamically hot structures and the motion of haloes within those structures, rather than the tidal field, is the most plausible driver of the passing-by motion and hence underlies the secondary bias of high- haloes.
While the above two regimes do not represent the entire halo population, they bracket the flow pattern and the driving forces at the two extremes, and the general halo population is expected to be a mixture of the two cases. In an evolutionary sense, a CDM halo may start as a local density peak embedded in a coherent flow, mimicking the condition expected for the cold regime, and then gradually fall into a dynamically hot, massive structure, mimicking the condition expected for the hot regime; during this evolution, the two driving forces, and , therefore compete or synergize with each other in shaping the flow pattern around and the accretion onto the halo, and dominate the secondary bias when the halo is in the corresponding regime. The switching of the dominant factor of the secondary bias, and the accompanying increase in the dynamical hotness of the embedding environment, are therefore a reflection of the hierarchical nature of structure formation in CDM cosmology, while our proposed indicator, , is a sensitive tracer of this formation process.
The two regimes, and the evolutionary connection between them, also imply a mass dependence of the dominant factor of secondary bias. A low-mass halo can start in a void and complete the migration to a massive filament or node, so that it can complete the switch from the cold to the hot regime, and from to as the dominant factor. In contrast, a massive halo (e.g. Figure 11c), especially an extremely massive one, is not likely to be embedded in a more massive structure, since these structures do not have enough time to assemble in the finite age of the Universe; the migration of such a halo to the hot regime is therefore incomplete, and the accretion onto the halo cannot be truncated by the inertial force, as indeed seen from the mass dependence of the divergence of CHM flow (Figure 11, Figure 12).
We note that the two distinct regimes are physical equivalents to the two scenarios, respectively, that are proposed in the literature to explain the secondary bias. The subdominance of tides in the high- case is equivalent to the gravitational/dynamical ‘heating’ scenario proposed and examined by e.g. Mo et al. (2005), Wang et al. (2007), and Mansfield & Kravtsov (2020), in which the heating of the CHM to high ‘temperature’ by the formation of large-scale structures is inferred to suppress the accretion onto haloes. In particular, Figure 8 of Wang et al. (2007) shows that particles surrounding older, low-mass haloes have larger velocities relative to the halo, often exceeding its virial velocity, whereas the corresponding velocity offsets around massive haloes are smaller. This provides a connection between the availability of slowly moving, capturable material and the formation-time dependence of halo clustering. In contrast, the dominance of tides in the low- case is equivalent to the scenario of tide-limited accretion proposed and discussed by e.g. Wang et al. (2011), Hahn et al. (2009), and Ramakrishnan et al. (2019), in which the tidal shear is found to correlate with halo properties. Our force-based formulation thus provides a unifying framework, as well as an evolutionary view, that connects the two scenarios.
The distinction between the two regimes also suggests why simple tidal-field metrics, such as the tidal anisotropy (), the classical Hill radius (; see e.g. Eq. (10) of Hahn et al. (2009)) and analogous formulations, may not fully account for the secondary bias of haloes (Mansfield & Kravtsov (2020)), since the halo-centric frame is not inertial, especially for the high- haloes, and the CHM acted on by the inertial force is not attached to the halo. In particular, the subdominance of the tidal force in the high- haloes implies that the reduction of residual secondary bias after controlling for tidal metrics, such as (Figure 7) and (Mansfield & Kravtsov (2020)), does not necessarily indicate that the tidal field is the factor that regulates halo accretion; instead, these tidal metrics serve as classifiers of the cosmic-web environment, with a highly anisotropic tidal field signaling the formation of a massive, dynamically hot, filamentary structure, which in turn generates randomized motion of embedded haloes and consequently the inertial forces that suppress the accretion. This agrees with Borzyszkowski et al. (2017), who show that for haloes embedded within thick filaments, the infalling material joins the filament rather than the halo, so that accretion onto these haloes is suppressed; in our force formulation, it is the inertial force, and the passing-by motion of the CHM it induces, that is responsible for such suppression. In Appendix B and Figure B2, we show that the CHM flow pattern is indeed distinct between haloes with different , providing further support for the classifier role of the tidal metrics.
Figure B2. Stacked velocity fields of haloes split by tidal anisotropy . As in Figure 11, the panels show the velocity fields around haloes split by in bins of halo mass.
The two-regime scenario not only explains the mediator role of in producing the secondary bias, but also provides a route to understanding the essential characteristics in the secondary bias of different internal halo properties. For the spin bias, the inertial force in the hot regime drives divergent flow around low-mass haloes, which prevents the high-spin material in the CHM from being delivered into the halo, as shown in Figure 12(a), (Figure 12d) and discussed in § 4.1.1, and explains the inverted bias at low mass (Figure 4a). The incomplete migration of massive haloes to the hot regime, as discussed above, limits the inertial-force-driven divergence, which allows the high-spin material to be delivered into the halo, and explains the increasingly positive bias at high mass (Figure 4a). The reduced inflow around high-, low mass haloes also provides an interpretation of their earlier formation and higher concentration (Figure 4a, Figure 4b): suppression of late accretion can leave a larger fraction of the final mass assembled at early times, resulting in a larger , a steeper density profile and a higher . Such a difference in the accretion history also leaves signatures in the outer boundaries of haloes, as demonstrated in Appendix C. The mass dependence of the bias and bias, however, is more complex, and we refer the reader to Appendix § 9 for a detailed interpretation.
fun tracing competitive accretion
The flow patterns in §4.1 show how the large-scale environment regulates the motion and boundness of the CHM. On smaller scales, a close companion can also modify the material available to a halo. The splashback population in § 3.4 illustrates such an effect in the presence of a past encounter with a more massive system. Here we examine the current configuration of the surrounding matter, particularly when two haloes compete for a common supply.
To characterize the local geometry that cannot be fully specified by the smoothed tidal anisotropy, , we use the environmental axis ratio, , defined in § 2.2 from all particles in the shell with a halo-centric radius of –. A prominent companion halo or a nearby filamentary structure increases the extent of the mass distribution along its major axis, producing a small , while a more nearly spherical distribution has a larger . Thus, can distinguish local configurations within a given tidal environment at large scale. Since the companion and the matter surrounding the pair also contribute to the local potential, this geometrical distinction can affect as well as the direction of accretion.
Figure 14 compares the ambient density and velocity fields of haloes in the upper and lower quartiles of . Within each mass bin, the two samples are reweighted to a common distribution, so their differences cannot be attributed simply to different distributions of tidal anisotropy. In the low- stacks (panels d–f), a prominent off-centre density concentration appears along the positive major axis, whose sign is chosen to point towards the denser side of the environment. The high- stacks (panels a–c) lack such a pronounced companion. At the low-mass end, the characteristic configuration is a halo beside a comparable or more massive neighbour, while at the high-mass end, the target halo is the more massive member of the pair. The high-mass, low- stack also shows an elongated distribution extending to the opposite side of the halo, indicating that the pair is embedded in a larger, filamentary structure.

Figure 14. Stacked velocity fields around haloes with different environmental axis ratio . The figure is similar to Figure 11, but the division of the halo sample is based on the axis ratio, , of the environment at halo-centric distances of 2–4 . To control the effect of tidal anisotropy , stacking is performed by weighting the haloes so that the distributions of for the haloes in the upper and lower quartiles of are aligned to the same distribution as that of all haloes in the same mass bin. To reveal the companion structures, the alignment of the halo-centric frame is chosen so that the major and intermediate axes of the inertia tensor involved in the computation of point to the and directions within the page, with requirements that the denser side outside points to the direction and that the spin vector using all particles within points out of the page to break the remaining degrees of freedom of reflection. In each panel, the greyscale background shows the median surface mass density in the same slice. This figure shows that haloes with lower at fixed tend to have a more prominent companion, which enhances the binding of the CHM but also competes for the accreted material (see § 4.3 for details).
A companion has two effects on the surrounding material. Together with the matter embedding the pair, it can deepen the local potential and increase the fraction of particles classified as bound. This contribution is included in our energy estimate, which uses all particles within (§ 2.2), and can therefore lower . At the same time, the companion attracts part of the common supply and limits the region from which the target halo can accrete. This competition is particularly restrictive for a low-mass halo beside a more massive neighbour, consistent with the tidal suppression of halo growth examined by Hahn et al. (2009). A low in such a configuration therefore indicates substantial binding in the potential shaped by the halo and its CHM, but it does not assign all of the CHM to the future growth of the halo due to the competition of the companion.
The relative importance of these two effects changes with halo mass. When the target halo is the dominant member of the pair, it can retain a larger share of the surrounding material, and the deeper potential is more readily accompanied by inflow towards its own centre. This expectation is consistent with the high-mass stacks: the low- sample shows stronger converging motion around the halo, including on the companion-facing side, than the high- sample (Figure 14f, Figure 14c). For low-mass haloes, the low- flow is more asymmetric and more restricted in space, and has lower velocity than the flow around more massive haloes. Competitive accretion therefore suggests a stronger dependence of on at high halo mass, where the target can more effectively capture material from the shared environment.
Figure 15 allows us to examine this expectation while separating the effects of and . The population contours show a positive – relation at low mass, a weaker dependence at intermediate mass, and a negative relation at high mass. A possible geometrical explanation is that a low-mass halo can be embedded within a much larger anisotropic structure, while its relatively small CHM shell samples a locally rounder distribution. Such a configuration is consistent with the dynamically hot, passing-by CHM described in §4.1. For a massive halo, the CHM extends over scales comparable to the surrounding structure and exhibits its elongation, so a larger can coincide with a smaller . This interpretation is also consistent with the weaker passing-by motion, and hence the smaller role of the inertial force in massive haloes, as formulated in §4.1.
Figure 15. Unbound fraction and bias as functions of tidal anisotropy and environmental axis ratio . (a)–(c), distribution of individual haloes (dots) within bins of in the plane of and , coloured by . (d)–(f), similar but coloured by . In each panel, black contours encompass 25, 50 and 75% of the haloes. The Loess algorithm (Cleveland 1979; Cappellari et al. 2013) is used to smooth the colored property to reveal the trend. This figure demonstrates that depends on environmental metrics at different scales, and that traces in the combined space of and (see § 4.3 for details).
The colour gradients in Figure 15(a)–(Figure 15c) show that generally increases with within each mass bin. At a given , also increases with , with a more pronounced dependence in the high-mass sample. Thus, the elongated configurations selected by small contain a larger bound fraction even when the smoothed tidal anisotropy is held approximately fixed. The stronger conditional trend at high mass is consistent with the greater ability of the dominant halo to retain the shared material. These conditional trends are thus consistent with competitive accretion contributing to the scatter in at fixed tidal anisotropy, and indicate that is sensitive to the environments at different scales for which and are proxies.
Figure 15(d)–(Figure 15f) show that the colour patterns of halo bias are similar to those of in the – plane. The density maps in Figure 14 offer a possible explanation for this additional dependence of on : the high- stack has a rounder distribution within , and lacks a prominent companion at this scale; however, it exhibits an extended overdensity at , indicating the presence of a massive, larger-scale structure beyond the CHM, and yielding a higher bias than the low- counterpart at fixed . The similarity between the dependence of and on at fixed is also explained: in order to generate a similar , the surrounding matter can either be a massive, large-scale overdensity, which yields a large and a small , or a less massive companion, which elevates within the CHM but reduces at large scales. Mansfield & Kravtsov (2020). Tracing particle trajectories would provide a direct test of how the common supply is divided between the two haloes.
Our interpretation suggests that the configuration of matter on different scales affects the accretion onto a halo, with a major proxy for that configuration and a complementary one. Indeed, Mansfield & Kravtsov (2020) found that a single-neighbour tidal proxy does not account for all low-mass assembly bias, and that tides from the extended matter distribution and the velocities of ambient particles must also be considered. Therefore, the combination of and is expected to account for much of the secondary bias, in a manner similar to , as seen from the similar gradients of and in the – plane (Figure 15). A limitation of our analysis and interpretation is that the instantaneous energy classification and stacked projected flow do not determine which member of a pair ultimately accretes a particle. Binding in halo outskirts is sensitive to the surrounding potential and tidal field (see appendix D of Mansfield & Kravtsov (2020)).
Summary and discussion
In this paper, we have used the TNG300-1-Dark simulation to investigate how the binding state of the circum-halo medium (CHM) is related to halo secondary bias. Our sample consists of haloes with – at . The unbound mass fraction, , measured in the 2–3 shell around each halo, has a strong correlation with halo bias, making it a sensitive indicator of secondary bias. Our main findings are as follows.
(i) Haloes with larger are more strongly clustered at every mass examined, with a correlation coefficient –0.6. Conditioning on reduces the correlation coefficients between bias and internal properties of haloes (formation time, concentration, spins using bound or all particles, shape, or velocity anisotropy) to a few percent. The reduction is nearly complete for spins and velocity anisotropy. Formation time, concentration, and shape only retain weak residual biases (§ 3.1).
(ii) We examine other environmental indicators that have been proposed to explain the origin of secondary bias. Among them, tidal anisotropy is strongly correlated with , and both substantially reduce secondary bias; leaves smaller residuals for spin and velocity anisotropy, while the relative performance depends on halo property and mass. Low-mass splashback haloes have elevated , and conditioning on reduces secondary bias within both the splashback and non-splashback samples. Weak residual biases in properties related to assembly history remain (§ 3.2).
(iii) The stacked velocity field and radial profiles reveal two distinct flow patterns for the CHM around haloes with the lowest and highest : one in which haloes co-move with their CHM, and the other in which haloes move through a dynamically hot, partially virialized environment. As a result, the CHM around low- haloes shows converging motion (more negative , smaller ), while that around high- haloes exhibits stronger passing-by flow (less negative or positive , larger ). The flow pattern not only determines the ability of the halo to accrete from the CHM, but also affects the delivery of angular momentum through the accretion. The distinction depends on halo mass and is most pronounced at low mass. At all masses, larger accompanies reduced inflow.
(iv) The physical mechanism producing secondary bias is thus two-fold, with distinct driving forces in the two regimes: in the ‘cold’ regime, tide-regulated CHM motion operates for low- haloes that co-move with their CHM embedded in dynamically cold structures at large scales; in the ‘hot’ regime, tidal force is subdominant, and the inertial force generated by acceleration of haloes operates for high- haloes moving through massive, dynamically hot structure. As a halo evolves, the hierarchical nature of structure formation in CDM cosmology implies that the dominant mechanism can switch from the former to the latter (§4.1).
(v) The local configuration of the CHM supplements information beyond the smoothed tidal anisotropy, and points to a competitive accretion scenario. At fixed , increases with the environmental axis ratio measured over –, most strongly at high mass. Stacks matched in associate small with elongated configurations containing a prominent companion or filamentary structure. The population-level – relation changes from positive at low mass to negative at high mass. A companion can deepen the potential and lower while competing for the share of the common supply to the target halo (§ 4.3).
The physical picture suggested by our results posits two regimes, at the extremes of a continuous spectrum of halo–environment interactions, in which inflow is regulated by tidal or inertial forces. Our statistical results support as a common mediator that traces both types of interaction.
Our picture also offers a physical interpretation of the adopted aperture. The – shell brackets two characteristic boundary scales (Appendix C): the splashback radius, the first apocentre of recently accreted material, is located at – for typical haloes (e.g. Diemer & Kravtsov 2014; More et al. 2015), while the depletion radius, out to which halo growth has exhausted the surrounding matter, typically lies at (e.g. Fong & Han 2021; Han 2026). The aperture therefore samples the transition zone just outside the splashback shell but within the depleted region, where the material is most sensitive to the balance between the halo potential and environment, and thus best encodes the halo–environment interaction. We caution that small residual secondary bias conditioned on does not in general prove conditional independence, and matching present-day does not erase differences in past accretion or encounters. The non-monotonic concentration trends at high mass (Appendix § 9) further caution against assigning a single growth history to each unbound fraction. Following haloes and particles across snapshots would test whether changes in precede changes in accretion, spin, and orbital structure, and would distinguish the material captured by a halo from that captured by its neighbours.
Energy definition and radial aperture also affect the interpretation of . We have verified that extending the potential aperture to does not change the main results. Nevertheless, the potential includes surrounding structures, and its reference radius is finite, so that is an operational measure of present-day binding, rather than a count of particles destined to escape to infinity (e.g. Behroozi et al. 2013; Mansfield & Kravtsov 2020). Similarly, median radial velocities diagnose typical inflow or outflow, not a mass accretion rate. Tests with alternative potential conventions, numerical resolutions, redshifts, and cosmologies are needed before treating the preferred shell or the measured correlations as universal.
The implications for galaxies require a further modelling effort. If reduced dark-matter inflow accompanies reduced gas accretion, galaxies in high- haloes could exhaust their gas reservoirs earlier. However, cooling, feedback, and interactions with the ambient gas may alter both the flow patterns and the transfer of angular momentum. These processes may be particularly violent for low-mass, high- haloes, around which the passing-by CHM has velocities comparable to or greater than (Figure 11a), and transonic or supersonic motion may arise for gas near the virial temperature. In this case, turbulent mixing between halo gas and the passing-by CHM may promote entrainment of ambient gas rather than the removal of halo gas (e.g. [19] (Li et al. 2023)). Such entrainment could transfer angular momentum from the CHM to the halo gas and potentially lead to larger galaxy sizes, providing a possible physical connection to the strong clustering of diffuse galaxies ([46] (Zhang et al. 2025)). A decoupling of gas and dark-matter spin has been reported on galactic scales in simulations (e.g. Velliscig et al. 2015; Jiang et al. 2019; Yang et al. 2023). The flow patterns found here suggest that such a decoupling may also arise on CHM scales, particularly around low-mass, high- haloes. Matched haloes in dark-matter-only and hydrodynamical simulations could test this possibility and establish whether predicts gas supply, star formation, or galaxy size.
Observational tests will require calibrated proxies for . Tidal anisotropy has been estimated for SDSS groups on a fixed scale of (Paranjape et al. 2018b), providing information on the halo environment. Complementary kinematic information may be obtained from neighbouring galaxies, whose coherent motions have been used to construct a halo-spin proxy (e.g. Kim et al. 2025). Such measurements suggest a possible route to probing circum-halo motions and, after calibration, the binding state traced by . The outer density profile offers a complementary diagnostic, given its dependence on (Appendix C). Combining these environmental, kinematic, and density measurements in mock catalogues would test whether they can trace and recover its relation to clustering at fixed halo mass, accounting for smoothing scale, projection, and sample selection.
Another approach is to incorporate directly into models of the galaxy–halo connection. It could serve as a secondary variable in a decorated halo occupation model (e.g. Hearin et al. 2016). Alternatively, an effective halo mass could be constructed from and , extending the approach of Xu & Zheng (2018). This would turn the environmental dependence identified here into a testable model of galaxy clustering.
Data availability
TNG300-1-Dark is available at https://www.tng-project.org.
Acknowledgements
Qinglin Ma thanks Cheng Li, Dandan Xu, Yifeng Zhou, Joop Schaye, Fangzhou Jiang and Jiaxin Han for helpful discussions. The computations and presentations in this paper are supported by various software: Scipy (Virtanen et al. 2020), Numpy ([87]), Matplotlib (Hunter 2007), Ipython (Perez & Granger 2007) and Astropy (Robitaille et al. 2013; [2, 3]). The authors acknowledge the use of Codex (OpenAI) as a writing and research aid in the preparation of this manuscript, including literature searches, drafting assistance, and language editing; all scientific content, analysis, and conclusions are the authors’ own.
The selection of the radial range of unbound particles
In the main text, our default choice of the radial range to measure is –, where its correlation with halo bias is strongest (Figure 3c). Here we justify this choice by repeating the correlation analyses of § 3.1 for three alternative radial ranges, –, –, and –. In Figure A1, we show the Spearman correlation coefficients between and each internal halo property (panels a–d), and the corresponding conditional correlation coefficients by controlling for (panels e–h), with each column adopting a different radial range to evaluate .
Figure A1. Spearman correlation coefficients using measured in different radial ranges. Each column shows the results similar to Figure 4 but using measured in a given radial range, with the upper row for the unconditional correlation coefficient between and each halo property , and the lower row for the correlation coefficient between each halo property and bias factor conditioned on .
The innermost shell, –, behaves differently from the others (Figure A1a, Figure A1e). Within the virial radius, the unbound particles are rare (; see Figure 2), and including them in the halo definition directly biases the measurements of the internal properties themselves: the inferred concentration is depressed because the unbound particles preferentially reside in the outer regions (see also (Ma & Li 2026)), while the spin is enhanced and the velocity anisotropy reduced, consistent with their larger tangential velocities and angular momenta. In this aperture, therefore, largely traces the contamination of the internal-property measurements rather than the halo environment, and its ability to account for the secondary bias is limited (Figure A1e).
For the outer shells, the correlations revert to the environmental pattern seen in the main text: in low-mass haloes, a higher is associated with a higher concentration and an earlier formation time (Figure A1b–Figure A1d). Among the four apertures, the (2–3) shell yields the strongest correlation with halo bias and, after conditioning on , the smallest residual secondary bias (Figure A1c, Figure A1g), consistent with the pronounced separation of the – relations in Figure 3c. The (1–2) shell gives intermediate results, while the (3–4) shell remains correlated with halo bias but leaves larger residuals for some properties (Figure A1d, Figure A1h).
We conclude that the choice of 2–3 is empirically optimal within the radial ranges examined here. A plausible physical origin of this aperture is discussed in §5: the shell brackets the transition zone between the splashback and depletion radii, where the boundedness of the surrounding material may be most sensitive to the interaction between the halo and its environment.
The role of tidal anisotropy in secondary bias
In §3.2.1, we showed that the tidal anisotropy (see §2.2 for the definition) is strongly correlated with , and that controlling for suppresses the secondary bias. Here we complement the correlation analysis by examining the correlation between and the other halo properties, and supplement the flow analysis of §4.1 by examining the CHM flows of haloes split by .
In Figure B1, we show the Spearman correlation coefficients between and each internal halo property . Consistent with the expectation (i) of §3.2.1, the correlations are significant over the entire mass range examined, and their signs and mass dependences closely mirror those of (Figure 4b): haloes with higher are older, more concentrated, and rounder at low mass, and exhibit more tangentially biased orbits. The conditional correlations after conditioning
on are shown in Figure 7 of the main text, where conditioning on is found to largely suppress the secondary bias, in agreement with Ramakrishnan et al. (2019). Quantitative discrepancies remain, however. First, we measure a stronger correlation between and the bias factor across the entire mass range (black curve in Figure B1), as well as stronger correlations between the secondary properties and (Figure 4a), than reported by Ramakrishnan & Paranjape (2020). This difference likely stems from the bias estimator: we measure the halo bias on intermediate scales (–), whereas Ramakrishnan & Paranjape (2020) adopt a power-spectrum method probing larger scales (); measuring the bias on such large scales may partially dilute the intrinsic correlation between and halo bias. Second, our most notable discrepancy lies in : both the – and – relations exhibit a pronounced mass dependence, whereas Ramakrishnan et al. (2019) observe almost no mass dependence. This discrepancy is likely attributable to the definition of : we consider only bound particles within , whereas Ramakrishnan & Paranjape (2020) include all particles inside the mass ellipsoid. Consequently, constrains the secondary bias slightly differently in the two studies. Nevertheless, our findings reinforce their primary conclusion that plays a pivotal role in the secondary bias.
In §4.1, we showed that the two flow patterns of the circum-halo medium are traced by . Here we show that the same dichotomy emerges when haloes are split by the tidal anisotropy . Figure B2 presents the stacked velocity fields constructed in the same way as Figure 11, but with haloes split within each mass bin by : the upper and lower rows show haloes in the top and bottom 25% of (Figure B2a–Figure B2c and d–f, respectively). The stacks support the expectation of §4.1: haloes in highly anisotropic environments have typically entered the vicinity of dynamically hot, massive structures, and the resulting inertial force dominates the motion of their CHM, driving the passing-by flow elevating ; in contrast, haloes in more isotropic environments co-move with their CHM, and the tidal force governs their relative motion, shaping converging, infall-dominated flows and lowering . This agreement between the -split and -split flows
underlies the strong correlation between the two quantities found in §3.2.1, and supports the interpretation that the tide is an indicator of the cosmic-web environment, rather than a direct driver of the CHM motion.
The imprint of unbound mass fraction on the halo boundary
In §4.1, we showed that high- haloes experience reduced inflow relative to low- haloes of the same mass. The growth state of a halo has been found to relate to the outer density profile and to the boundary radii: rapidly accreting haloes develop steep outer profiles whose logarithmic-slope minimum marks the splashback radius, the first apocentre of recently accreted material (e.g. Diemer & Kravtsov 2014; More et al. 2015; Shi 2016), while the growth of the halo depletes the surrounding matter out to a depletion radius of , revealed as a minimum of the halo–matter bias profile (e.g. Fong & Han 2021; Han 2026). It is thus expected that high- and low- haloes should exhibit systematically different boundary structures. To test this, in Figure C1, we show the density profile (a–c), its logarithmic slope (d–f), and the bias profile (g–i; see §4.5 for the definition), for haloes in the same mass bins and quartiles as in Figure 12.
Figure C1. Radial profiles of density metrics around haloes. The density metrics shown are the density contrast (panels a–c), its logarithmic slope (panels d–f), and the bias factor (panels g–i; see Eq. (5) for the definition). Each column shows the results for haloes in a bin of . Profiles in panels (a)–(c) are median values among haloes in use, and in panels (d)–(f) are the logarithmic slopes of the corresponding median density profiles, and in panels (g)–(i) are the average value of the bias factor. In each panel, coloured curves show those profiles for haloes in different quartiles of , while black curve shows that for all haloes, with a vertical dashed line marking the median virial radius of these haloes.
At low mass, the high- haloes exhibit a higher concentration, although the difference is subtle in this representation (Figure C1a–Figure C1c). More evident is that high- haloes show an enhanced density around the halo boundary, consistent with the ambient material associated with massive host structures streaming past the haloes; moreover, both the splashback and the depletion radii, marked by the minima of the logarithmic slope and of the bias profile, respectively (Figure C1d–Figure C1f, g–i), are located at smaller radii. Low- haloes, which are still gaining mass, show the opposite behaviour: ongoing accretion enlarges the region over which the halo shapes its surroundings, pushing both boundary radii outward, consistent with Diemer & Kravtsov (e.g. 2014). This boundary difference provides a potential observational test of the –bias connection, since the outer density profile can be measured through weak lensing or galaxy counts.
Interpreting secondary bias across halo mass
In § 3.1, we showed that the secondary bias of the various internal properties differs in strength and even in sign. This appendix examines how the six internal properties (see § 2.2 for the definitions) are related to the two complementary environmental metrics, and (Figure D1), and uses these relations to interpret the characteristics of the secondary bias of different internal halo properties across halo mass.

Figure D1. Internal properties in the – plane. The panels show individual haloes, colour-coded by formation time (), concentration (), spins in two definitions (; ), shape () and velocity anisotropy () (from top to bottom), in three mass bins (from left to right). The contours encompass 25, 50 and 75% of the haloes in the – plane.
The six properties fall into three categories:
(i) The assembly metrics: formation time () and concentration (). At low mass, both and increase with and . Their
gradients are similar to those of halo bias and in the – plane (Figure 15d), consistent with the increased suppression of late accretion towards the ‘hot’ regime or the lack of close neighbours to deepen the binding potential, both preferring high- environments at large scales (§ 4). At high mass, however, a new population emerges at the lower-left corner in the – plane, in which both and are high but is low (see Figure 15d). A possible origin for this population is that a massive halo without an embedded overdensity (such as a filament or node) may have exhausted its surroundings and stopped accreting, as indeed seen in the low bias profile and large depletion radius of the high-mass, low- haloes (Figure C1). A selection of high haloes thus includes those in both the upper-right (high ) and lower-left (low ) corners of the – plane, leading to a cancellation of and thus weakened secondary bias of at high mass. The lower-left population is more prominent in than in , indicating that the concentration is more sensitive to the exhausted accretion; a direct consequence is that a selection of high haloes includes more low- haloes, which negates the average and reverses the secondary bias of at high mass.
(ii) Spins, and , which are defined using all particles and bound particles, respectively. The consistent sign of – across halo mass, and the inverted sign of the – correlation at the low masses (Figure 4a), as already discussed in §4.1, are owing to a combination of two factors that emerge in the CHM pattern at high : the enhanced vorticity, which sources the angular momentum of CHM, and the reduced negative divergence, which reduces the delivery of angular momentum from CHM to the halo.
(iii) The anisotropy metrics: shape () and the velocity anisotropy (). Both exhibit a monotonic gradient in the – plane, aligning with the gradient of and . The mass-invariant sign of the – and – correlation (Figure 4a) indicates that these metrics are insensitive to the accretion rate, but rather reflect accretion geometry. The interpretation can also be made by our two-regime scenario and competitive accretion picture: the passing-by flow in a ‘hotter’ (larger ) environment produces larger vorticity, and thus less radially biased orbits (smaller ) and less elongation along the accretion direction (larger ); the presence of a companion (lower ) in contrast, enhances the accretion via competitive accretion, producing more radially biased orbits (larger ) and more elongation along the accretion direction (smaller ).
This paper has been typeset from a TeX/LaTeX file prepared by the author.
References
- [2]Astropy Collaboration et al., 2018, AJ, 156, 123
- [3]Astropy Collaboration et al., 2022, ApJ, 935, 167
- [19]Du M., et al., 2019, ApJ, 875, 58
- [25]Gill S. P. D., Knebe A., Gibson B. K., 2005, MNRAS, 356, 1327
- [46]Ma C., et al., 2026a, Faintest Population Pattern of Baryon Cycle Revealed in Galaxy Ecosystems, doi:10.48550/arXiv.2609.21374, https://ui.adsabs.harvard.edu/abs/2026arXiv260921374MDOI
- [52]Mo H. J., White S. D. M., 1996, MNRAS, 282, 347
- [54]Mo H., van den Bosch F., White S., 2010, Galaxy Formation and Evolution. Cambridge University Press
- [69]Planck Collaboration et al., 2016, A&A, 594, A13
- [87]Van Der Walt S., Colbert S. C., Varoquaux G., 2011, Computing in Science and Engineering, 13, 22
- [102]White S. D. M., Rees M. J., 1978, MNRAS, 183, 341















