betaFeedback

WorksPhysics and Astronomy

Predicting Individual Dark Matter Halo Concentrations from Their Assembly Histories

Abstract

We present an empirical model for predicting concentration parameters of individual dark matter halos in cold and warm dark matter (WDM) cosmologies over z=0-5. Constructed using a large suite of cosmological N-body simulations, our model separately incorporates cosmic expansion (linear growth factor), mass scale (linear density fluctuations), halo mass assembly histories via the Extended Press-Schechter (EPS) scaled formation variable, plus a WDM damping term tied to the half-mode mass. It works for both virial and 200 ρcritρ_{crit} halo definitions. Across all tested cosmologies and redshifts, our global logarithmic fit gives concentration scatter ~0.09 dex for relaxed halos and ~0.09 to 0.10 dex for all halos. While relaxed halos maintain nearly constant scatter near 0.09 dex, the scatter of all halos at z>3 increases from 0.10 to 0.18 dex, revealing substantial violent-relaxation impacts on high-redshift halo structure. At fixed mass, concentration scatter is dominated by assembly history. We extract the scaled formation variable's probability distribution function (PDF) from merger trees and find departures from standard EPS theory. This PDF is narrower in WDM than LCDM due to suppressed small-scale perturbations and rapid, delayed monolithic collapse near the WDM truncation scale. Monte Carlo realizations with our formulae reproduce concentration distributions and match our prior peak-concentration model, whereas standard EPS underestimates concentrations from delayed-formation bias. Our framework offers a transparent, computationally lightweight tool to predict individual halo concentrations across diverse cosmological models.

Introduction

The concentration parameter is a fundamental descriptor of the internal structure of dark matter halos. Within the standard NFW ([31, 32]) description of halo density profiles, it characterizes the relative size of the halo boundary and the inner scale radius, and therefore provides a compact measure of how centrally condensed a halo is. Concentration is widely used in studies of nonlinear structure formation, the galaxy–halo connection, gravitational lensing, and dark matter phenomenology.

A large number of studies have established that halo concentration depends systematically on halo mass, redshift, and cosmology (e.g., [19, 7, 14, 43, 45, 36, 28, 34, 15, 11, 46, 26, 25, 12, 10, 9, 18, 42, 41 hereafter Paper I]). Most of this work, however, has focused on population-level statistics, such as the mean or median concentration–mass relation. In our previous work ((Paper I)), we argued that the peak of the concentration distribution provides a robust characterization of the most typical halo population. By extending the conventional peak-height framework to include information from halo mass accretion histories (MAH), we constructed a formation peak-height parameter and used it to derive a universal fitting formulae for the peak concentration over a broad range of cosmologies and redshifts.

Corresponding author: Weipeng Lin [email protected]

Corresponding author: Dao-Zhou Wang [email protected]

A complementary question is how to describe the halo-to-halo scatter at fixed mass and redshift. This issue is physically important because halos with the same virial mass observed at the same epoch can still show substantial structural diversity. Such scatter directly affects applications that depend on halo internal structure. In the studies of galaxy formation and evolution, concentration governs the central gravitational potential well and inner circular velocity profile, thereby modulating gas cooling, galaxy sizes, rotation curves, and assembly bias in the galaxy–halo connection. In large-scale structure analyses, capturing halo-to-halo structural variations is essential for modeling small-scale nonlinear clustering, predicting gravitational lensing signals, and constructing realistic mock catalogs for wide-field galaxy surveys.

A long-standing idea is that halo-to-halo scatter in concentration is closely tied to mass accretion history. Earlier studies showed that earlier-forming halos are generally more concentrated (e.g., [43, 47, 46]), indicating that concentration retains memory of halo assembly. A. D. Ludlow et al. [24] developed a physically motivated model that links the characteristic inner density to the critical density at the epoch when the corresponding mass scale first collapsed. This model reproduces the median concentration–mass relation and, given the complete collapse-mass history of all progenitors, also predicts individual-halo concentrations in both Λ\LambdaCDM and Λ\LambdaWDM cosmologies. More recently, T. Zhang et al. [44] showed with neural networks that MAH information substantially improves halo-by-halo concentration predictions, although their reported root mean square error (RMSE) in logarithmic concentration space increases from 0.0845 to 0.1287 over z=0z = 0–2. However, systematic tests and calibrations of individual halo concentration predictions beyond z=2z = 2 remain scarce. Nevertheless, these studies suggest that concentration scatter largely reflects the diversity of halo formation histories, while leaving open whether individual concentrations can be captured by a compact, physically interpretable, and nearly universal relation across redshift, cosmology, and halo definition.

This work aims to develop a simple and compact empirical model for the concentration parameter of individual halos. While Paper I encapsulated the mass and redshift dependence of the population peak concentration into an effective formation peak height, in this work we decouple the cosmological linear growth factor D(z)D(z) and the linear density rms fluctuation σ(M)\sigma(M) as independent regressors. This decoupled formulation provides enhanced flexibility to independently capture cosmic expansion background effects and initial power spectrum amplitude scalings. To parameterize the individual MAH of each simulated halo, we utilize the EPS-scaled formation variable ωf\omega_{\mathrm{f}} (C. Lacey & S. Cole [21], see below for definition). For WDM cosmologies, we further include the half-mode mass MhmM_{\mathrm{hm}}, which characterizes the cutoff scale in the linear power spectrum and is defined in detail in § 2.4. Using a broad suite of cosmological simulations, we primarily calibrate an empirical, compact relation for dynamically relaxed halos that connects individual halo concentrations to their collapse rarity, assembly history, and the WDM cutoff scale. We also provide the corresponding relation for the complete all-halo sample for comparison. We first verify this formulation through a direct halo-by-halo comparison with simulation measurements across redshifts z=0z = 0–5. We then demonstrate how the overall concentration distribution can be statistically reconstructed from the halo formation-time distribution via Monte Carlo realizations, showing excellent consistency with the peak-concentration relations established in Paper I. Finally, we compare our empirical framework against existing physically motivated and machine-learning models to highlight its robustness, computational efficiency, and physical interpretability across diverse cosmological scenarios.

This paper is organized as follows. § 2 describes the simulation samples, relaxed halo selection, and model variables. § 3 presents the empirical relation, tests it with EPS- and simulation-based formation times, and compares the predicted concentrations with simulations and previous individual halo models. § 4 summarizes our conclusions. Throughout this paper, ‘log’ denotes a base-10 logarithm.

Data and Method

Numerical simulations

The simulations used in this work are based on the data set of Paper I, supplemented by the Uchuu simulation suite. We use the dark-matter-only runs from the TNG project1 ([29, 30, 33, 35, 39]), the MultiDark-Planck simulations2 ([20]), and the Uchuu simulations3 ([18]) as our Λ\LambdaCDM samples. These public simulations provide a broad coverage in halo mass and numerical resolution, from low-mass halos resolved in small high-resolution boxes to massive cluster-scale halos sampled in large cosmological volumes.

To test the dependence on cosmology, we also include the non-Λ\LambdaCDM simulations described in Paper I. These simulations were performed specifically for our project and consist of an SCDM\mathrm{SCDM} model, an OCDM\mathrm{OCDM} model, a scale-free Einstein–de Sitter (EdS) model with an initial power spectrum P(k)∝k−2P(k) \propto k^{-2}, two wwCDM models, and a set of Λ\LambdaWDM models. The two wwCDM simulations adopt constant dark-energy equation-of-state parameters corresponding to the first and 37th models of the Coyote suite (see K. Heitmann et al. [16], K. Heitmann et al. [17] for details), while the Λ\LambdaWDM simulations use equivalent thermal relic particle masses of mWDM=0.5m_{\mathrm{WDM}} = 0.5, 0.8, 1.1, and 1.5 keV.

The main numerical settings and cosmological parameters of all simulations are summarized in Table 1. The full sample spans box sizes of 30–2500 h−1Mpch^{-1}\mathrm{Mpc} and particle masses of 3.65×1053.65 \times10^{5}–2.36×10102.36 \times10^{10} h−1M⊙h^{-1}M_{\odot}, allowing us to study halo concentrations over a wide range of masses, redshifts, resolutions, and cosmological models.

SimulationLboxL_{\mathrm{box}} (h−1Mpch^{-1}\mathrm{Mpc})NDMN_{\mathrm{DM}}mDMm_{\mathrm{DM}} (h−1M⊙h^{-1}M_{\odot})ϵ\epsilon (h−1kpch^{-1}\mathrm{kpc})Ωm\Omega_mΩΛ\Omega_{\Lambda}σ8\sigma_8nsn_shhww or mWDMm_{\mathrm{WDM}} (keV)Code
TNG50-dark35216032160^36.65×1056.65 \times10^50.200.30890.69110.81590.96670.6774⋯\cdotsAREPO
TNG100-dark75182031820^36.0×1066.0 \times10^60.500.30890.69110.81590.96670.6774⋯\cdotsAREPO
TNG300-dark205250032500^34.73×1074.73 \times10^71.00.30890.69110.81590.96670.6774⋯\cdotsAREPO
VSMDPL160384033840^36.20×1066.20 \times10^62.00.3070.6930.82280.9600.6777⋯\cdotsGADGET-2
SMDPL400384033840^39.63×1079.63 \times10^71.500.3070.6930.82280.9600.6777⋯\cdotsGADGET-2
MDPL21000384033840^31.51×1091.51 \times10^95.00.3070.6930.82280.9600.6777⋯\cdotsGADGET-2
BigMDPL2500384033840^32.36×10102.36 \times10^{10}10.00.3070.6930.82280.9600.6777⋯\cdotsGADGET-2
Micro-Uchuu1006403640^33.27×1083.27 \times10^84.270.30890.69110.81590.96670.6774⋯\cdotsGreeM
Shin-Uchuu140640036400^38.97×1058.97 \times10^50.400.30890.69110.81590.96670.6774⋯\cdotsGreeM
Mini-Uchuu400256032560^33.27×1083.27 \times10^84.270.30890.69110.81590.96670.6774⋯\cdotsGreeM
SCDM1005123512^32.07×1092.07 \times10^94.881.000.551.00.50⋯\cdotsGADGET-4
OCDM1005123512^36.20×1086.20 \times10^84.880.3001.01.00.667⋯\cdotsGADGET-4
SF-n21005123512^32.07×1092.07 \times10^94.881.001.0-2.0⋯\cdots⋯\cdotsGADGET-4
Λ\LambdaWDM-L100-0.51005123512^36.43×1086.43 \times10^84.880.310.691.01.00.67660.5GADGET-4
Λ\LambdaWDM-L100-0.81005123512^36.43×1086.43 \times10^84.880.310.691.01.00.67660.8GADGET-4
Λ\LambdaWDM-L100-1.11005123512^36.43×1086.43 \times10^84.880.310.691.01.00.67661.1GADGET-4
Λ\LambdaWDM-L100-1.51005123512^36.43×1086.43 \times10^84.880.310.691.01.00.67661.5GADGET-4
Λ\LambdaWDM-L60-1.5605123512^31.39×1081.39 \times10^82.930.310.691.01.00.67661.5GADGET-4
Λ\LambdaWDM-L30-1.530102431024^32.17×1062.17 \times10^60.730.310.691.01.00.67661.5GADGET-4
wwCDM11005123512^38.91×1088.91 \times10^84.880.43080.56920.81610.94680.5977-0.816SWIFT
wwCDM371005123512^35.79×1085.79 \times10^84.880.280.720.90001.02330.7313-1.294SWIFT

Table 1. The cosmological parameters and setting of simulations. The columns list the simulation identifier, comoving box size LboxL_{\mathrm{box}}, number of dark matter particles NDMN_{\mathrm{DM}}, particle mass mDMm_{\mathrm{DM}}, Plummer-equivalent gravitational softening length ϵ\epsilon, the cosmological parameters Ωm\Omega_m, ΩΛ\Omega_{\Lambda}, σ8\sigma_8, nsn_s, and hh, the additional model parameter ww for wwCDM cosmologies or mWDMm_{\mathrm{WDM}} for WDM cosmologies, and the simulation code.

Halo catalogs and concentration measurements

Halo catalogs and merger trees are taken from public releases whenever available. For the Λ\LambdaCDM cosmological simulations, we use the corresponding public halo catalogs and merger-tree products. For the simulations generated in this project, halos are identified with the phase-space halo finder Rockstar (P. S. Behroozi et al. [2]), and merger trees are constructed with Consistent trees (P. S. Behroozi et al. [3]).

Unless otherwise stated, our main analysis adopts the virial halo definition, for which the overdensity Δvir\Delta_{\mathrm{vir}} is computed using the fitting formula of G. L. Bryan & M. L. Norman [6] for each cosmology and redshift. More generally, for a halo defined by an overdensity threshold Δ\Delta, we denote its mass and radius by MΔM_{\Delta} and rΔr_{\Delta}, and define the concentration parameter as

cΔ≡rΔrs(1)c_{\Delta} \equiv\frac{r_{\Delta}}{r_{\mathrm{s}}} \tag*{(1)}

where rsr_{\mathrm{s}} is the NFW scale radius, often denoted by r−2r_{-2}, namely the radius at which the logarithmic slope of the density profile equals −2-2. Our fiducial results use Δ=Δvir\Delta= \Delta_{\mathrm{vir}} and are denoted by MvirM_{\mathrm{vir}}, rvirr_{\mathrm{vir}}, and cvirc_{\mathrm{vir}}. We also present results for Δ=200c\Delta= 200c, denoted by M200cM_{200c}, r200cr_{200c}, and c200cc_{200c}, in order to compare with previous work and to quantify the dependence of our concentration relations on the halo overdensity definition.

For the MultiDark-Planck, Uchuu, and our own simulations, halo concentrations are taken from the corresponding halo catalogs. The TNG public catalogs provide only c200cc_{200c}, whereas both cvirc_{\mathrm{vir}} and c200cc_{200c} are required in this work. We therefore derive the TNG concentrations by fitting NFW profiles directly, following the same procedure as in Paper I. This gives a consistent set of cvirc_{\mathrm{vir}} and c200cc_{200c} measurements for the TNG halos.

In the profile fitting, spherically averaged density profiles are constructed with 50 logarithmically spaced radial bins. For most simulations, the fits are performed over the radial range from 3ϵ3\epsilon to rvirr_{\mathrm{vir}}, where ϵ\epsilon is the Plummer-equivalent softening length. As shown in Paper I, 3ϵ3\epsilon is larger than or comparable to the convergence radius rconvr_{\mathrm{conv}} ([27]) for these simulations. For the TNG simulations, however, 3ϵ<rconv3\epsilon< r_{\mathrm{conv}} for a non-negligible fraction of halos, and we therefore set the inner fitting radius to rconvr_{\mathrm{conv}}. This conservative choice avoids the unresolved inner region and yields consistent NFW concentration estimates across the simulation set.

To reduce numerical-resolution effects and ensure robust concentration measurements, we retain only halos resolved by at least 2000 dark matter particles within rΔr_{\Delta}, consistent with the conservative criteria of H. L. Child et al. [8] and Paper I. Because the simulations have different mass resolutions, the corresponding minimum halo mass varies from one simulation to another; overlaps between simulations are used to check the consistency of the resulting concentration measurements.

In addition to the particle-number cut, we remove halos with unreliable merger histories or poorly resolved profile fits. We discard halos whose main progenitors were identified as satellite subhalos at any earlier snapshot, since these objects are often associated with flyby or backsplash events and can have artificially high concentrations, especially at low masses. This cut follows the spirit of D. Anbajagane et al. [1]. We also exclude low-mass halos for which the fitted NFW scale radius is comparable to the gravitational softening length, as their inner profiles are not sufficiently resolved and their fitted concentrations can be spuriously large. These cuts are applied before deriving the concentration relations below.

Relaxed halos

Because concentration is sensitive to the detailed shape of the halo density profile, we restrict the analysis primarily to relaxed halos. This helps remove systems that are strongly affected by recent mergers or other transient disturbances.

A halo is classified as relaxed if it satisfies

xoff<0.07,λ<0.07,2T/∣U∣<1.35,(2)x_{\mathrm{off}} < 0.07,\qquad\lambda< 0.07,\qquad2T/|U| < 1.35, \tag*{(2)}

where xoffx_{\mathrm{off}} is the offset between the halo center and center of mass, λ\lambda is the spin parameter, and TT and UU are the kinetic and potential energies. Following A. F. Neto et al. [34], A. D. Ludlow et al. [25], and A. Klypin et al. [20], we adopt the stricter virial-ratio cut 2T/∣U∣<1.352T/|U| < 1.35, which was introduced as a more conservative alternative to the commonly used 2T/∣U∣<1.52T/|U| < 1.5 criterion given the uncertainties in estimating the virial parameter.

The relaxed fraction decreases steadily with redshift, reflecting the fact that halos are dynamically younger and more frequently disturbed at earlier times. A similar redshift dependence of halo relaxation indicators was also found by A. Klypin et al. [20], who showed that the distributions of xoffx_{\mathrm{off}}, λ\lambda, and 2T/∣U∣−12T/|U| - 1 vary between z=0z = 0 and z=3z = 3. For the full sample, the relaxed fractions at z=0z = 0–5 are approximately 84%, 80%, 73%, 63%, 51%, and 34%, respectively. While our fiducial analysis focuses on relaxed halos, we also evaluate the complete, unselected halo sample (hereafter the “all-halo” sample) without applying dynamic relaxation cuts to evaluate the general applicability of our model across all evolutionary phases.

Model formulation and parameterization

Our central assumption is that the scatter in halo concentration is mainly driven by differences in halo formation history. The detailed procedures for computing D(z)D(z), σ(M)\sigma(M), and MhmM_{\mathrm{hm}} are broadly based on Paper I, with several key methodological refinements detailed below. In standard excursion set theory and halo models, the joint dependence on halo mass, redshift, and cosmology is often parameterized via the peak height ν(M,z)=δsc(z)/σ(M)=δsc(0)/[D(z)σ(M)]\nu(M,z) = \delta_{\mathrm{sc}}(z)/\sigma(M) = \delta_{\mathrm{sc}}(0)/[D(z)\sigma(M)], where δsc(0)=1.686\delta_{\mathrm{sc}}(0) = 1.686 and D(z)D(z) is the linear growth factor normalized to unity at z=0z = 0. The quantity σ(M)\sigma(M) is the linear density rms fluctuation on mass scale MM at z=0z = 0, computed from CAMB (A. Lewis et al. [22]). Rather than forcing D(z)D(z) and σ(M)\sigma(M) into a single composite variable ν\nu, in this work we decouple D(z)D(z) and σ(M)\sigma(M) as independent regressors, allowing the empirical relation to capture background expansion and mass scale scalings with maximum flexibility.

Following Paper I, the linear mass variance σ(M)\sigma(M) is computed using a real-space top-hat filter for CDM cosmologies. For WDM cosmologies, the sharp suppression of power below the free-streaming scale introduces artificial power leakage when evaluated with a real-space top-hat filter due to its k−3k^{-3} tail in Fourier space. To prevent this spurious small-scale contribution, we adopt a sharp-kk filter in Fourier space for WDM models (e.g., A. Schneider et al. [38]; Paper I).

To further encode the dependence on individual halo MAH, we utilize the EPS-scaled formation variable (C. Lacey & S. Cole 1993)

ωf=δsc(zf)−δsc(z)σ2(fM)−σ2(M)(3)\omega_f = \frac{\delta_{\mathrm{sc}}(z_f)-\delta_{\mathrm{sc}}(z)}{\sqrt{\sigma^2(fM)-\sigma^2(M)}} \tag*{(3)}

to serve as a proxy for halo formation epoch. Here zfz_f is defined as the redshift when the main progenitor first attains half (i.e. f=1/2f = 1/2) of the halo mass (MvirM_{\mathrm{vir}} or M200cM_{200\mathrm{c}}) at the redshift considered. In this work, zfz_f is measured from the main branch of the merger tree for each simulated halo. Compared with zfz_f, ωf\omega_f encodes implicit dependencies on mass, redshift, and cosmology within the excursion-set framework, thus yielding a more uniform description of halo assembly across different simulations.

For WDM cosmologies, the suppression of small-scale power introduces an additional physical scale. We characterize this effect using the half-mode mass MhmM_{\mathrm{hm}}, defined as the mass corresponding to the wavenumber khmk_{\mathrm{hm}} at which the WDM transfer function is reduced by a factor of 2 relative to the corresponding CDM model with the same background cosmological parameters: Mhm=4π3ρˉ(πkhm)3M_{\mathrm{hm}} = \frac{4\pi}{3}\bar{\rho}\left(\frac{\pi}{k_{\mathrm{hm}}}\right)^3. In the empirical model below, this cutoff scale enters through

Ahm=1+aMhmMΔ.(4)A_{\mathrm{hm}} = 1 + a\frac{M_{\mathrm{hm}}}{M_{\Delta}}. \tag*{(4)}

Prior to performing the multivariate linear regression, the parameter aa is fixed to the values calibrated from the peak-concentration model (listed in Table 2; see Appendix § 7 for details). For cosmologies without a WDM-like cutoff, Mhm=0M_{\mathrm{hm}} = 0 and hence Ahm=1A_{\mathrm{hm}} = 1.

SampleΔh=Δvir\Delta_{\mathrm{h}} = \Delta_{\mathrm{vir}}Δh=200c\Delta_{\mathrm{h}} = 200\mathrm{c}
AAαD\alpha_Dασ\alpha_\sigmaβ\betaγ\gammaaaσres\sigma_{\mathrm{res}}AAαD\alpha_Dασ\alpha_\sigmaβ\betaγ\gammaaaσres\sigma_{\mathrm{res}}
Relaxed halos0.611.030.660.63-0.454.770.0880.460.790.640.67-0.395.590.089
All halos0.591.050.630.68-0.433.980.0940.450.820.610.71-0.483.060.096

Table 2. Best-fitting coefficients of the individual halo concentration model (Equation (5)) across two halo boundary definitions, for relaxed halos and the complete all-halo sample. The cutoff coefficient aa in AhmA_{\mathrm{hm}} is fixed to the peak-model values.

To capture the joint dependence on cosmological expansion and power-spectrum fluctuation amplitude with maximum flexibility, we express the concentration of individual halos as

log⁡cΔ=A+αDlog⁡D(z)+ασlog⁡σ(M)+βlog⁡(1+ωf)+γlog⁡Ahm.(5)\begin{aligned} \log c_{\Delta} &= A + \alpha_D \log D(z) + \alpha_{\sigma} \log\sigma(M) \\ &\quad+ \beta\log(1+\omega_f) + \gamma\log A_{\mathrm{hm}}. \tag*{(5)} \end{aligned}

Here, the scaling coefficients αD\alpha_D, ασ\alpha_{\sigma}, β\beta, and γ\gamma capture the power-law dependencies on cosmic expansion, halo mass scale, assembly history, and the WDM cutoff suppression, respectively. We calibrate Equation (5) across the combined simulation suite via multiple linear regression, performed separately for relaxed halos and the complete all-halo sample.

We then test whether the concentration distribution can be predicted from the formation-time distribution. This is done in two ways. First, we combine Equation (5) with the EPS-predicted distribution of ωf\omega_f, which tests whether formation history scatter can explain concentration scatter statistically. Second, we replace the EPS prediction with ωf\omega_f measured directly from simulated merger trees, which isolates the empirical concentration model from possible inaccuracies in EPS.

Results

Universal relation for individual halos

We first test whether Equation (5) can universally describe individual halo concentrations from z=5z = 5 to z=0z = 0. Because halos across our diverse simulation suites—spanning different cosmologies, volumes, and resolutions—follow a broadly consistent sequence, we combine all samples to perform a global multiple linear regression for both the relaxed and all-halo populations.

The best-fitting coefficients of Equation (5) for all combinations of dynamical states and halo definitions are summarized in Table 2. Rather than writing out separate algebraic equations for each subset, Equation (5) serves as the master model relation, with specific parameters provided in Table 2. The fitted relations share consistent functional dependencies on background expansion, density fluctuation amplitude, and assembly history. The coefficients αD\alpha_D and ασ\alpha_{\sigma} are both positive, which is physically equivalent to a negative dependence on peak height (ν∝[Dσ]−1\nu\propto[D\sigma]^{-1}): halos with smaller mass (larger σ\sigma) and at lower redshift (larger DD) have systematically higher concentrations. The positive coefficient of log⁡(1+ωf)\log(1+\omega_f) demonstrates that earlier-forming halos are more concentrated, directly confirming that concentration retains memory of the early collapse phase.

Figure 1 compares the measured concentrations with the regression predictions for both relaxed halos and the complete all-halo sample. The top row presents the results for Δh=Δvir\Delta_h = \Delta_{\mathrm{vir}}, while the bottom row presents the results for Δh=200c\Delta_h = 200\mathrm{c}. Despite combining halos from diverse cosmologies, simulation volumes, resolutions, and redshifts, data points across all redshifts (z=0z = 0–55) cluster tightly around the one-to-one relation in all panels. This demonstrates that the physical parameters D(z)D(z), σ(M)\sigma(M), and ωf\omega_f successfully absorb the bulk of the explicit redshift, mass, and cosmological dependencies.

Comparison between measured and predicted concentrations for the combined simulation dataset

Figure 1. Comparison between measured and predicted concentrations for the combined simulation dataset. The top row shows virial concentrations (Δh=Δvir\Delta_{\mathrm{h}} = \Delta_{\mathrm{vir}}), while the bottom row shows 200c concentrations. The left column corresponds to relaxed halos, and the right column corresponds to the complete all-halo sample. Predictions, log⁡cΔmod\log c_{\Delta}^{\mathrm{mod}}, are calculated using Equation (5) with the best-fitting coefficients from Table 2. The distribution of individual halos is visualized via hexagonal binning color-coded by the mean redshift in each cell. Dashed lines indicate the one-to-one relation, while the contours enclose 95% and 68% of halos. Lower sub-panels display the binned dispersion, σlog⁡cΔ\sigma_{\log c_{\Delta}}, as a function of the model prediction.

As shown in the left-column panels of Figure 1, the relations for relaxed halos remain remarkably tight across all redshifts, with a stable residual scatter of σlog⁡c≈0.08\sigma_{\log c} \approx0.08–0.100.10 dex. In contrast, the complete all-halo sample (the panels in the right column) exhibits pronounced redshift evolution: its scatter is similarly narrow (∼0.09\sim0.09 dex) at z=0z = 0, but broadens significantly beyond z=3z = 3, reaching σlog⁡c≈0.18\sigma_{\log c} \approx0.18 dex by z=5z = 5. This elevated high-redshift dispersion is directly attributable to the dynamical state of the halo population discussed in § 2.3: as the relaxed fraction drops precipitously from 84% at z=0z = 0 to only 34% at z=5z = 5, approximately two-thirds of high-zz halos are dynamically disturbed and undergoing violent mergers or rapid non-equilibrium accretion. These unrelaxed systems produce substantial transient scatter in halo internal structures and NFW profile fits, thereby driving up the high-redshift dispersion. Furthermore, as shown in Table 2, the all-halo sample displays a slightly lower normalization AA, reflecting the well-known tendency of unrelaxed halos to possess lower concentrations at fixed mass.

The comparison between the two halo definitions indicates that the c200cc_{200\mathrm{c}} relation is systematically shifted to lower concentrations than the cvirc_{\mathrm{vir}} relation, consistent with the smaller normalizations in Table 2. This behavior is physically expected: r200cr_{200\mathrm{c}} probes an inner, denser region of the halo compared to rvirr_{\mathrm{vir}}, so M200c<MvirM_{200\mathrm{c}} < M_{\mathrm{vir}} and r200c<rvirr_{200\mathrm{c}} < r_{\mathrm{vir}}, yielding lower concentrations and a slightly modified sensitivity to background expansion.

Ultimately, Equation (5) provides a simple empirical description of individual halo concentrations. An individual halo’s concentration is primarily governed by three fundamental factors: cosmic background expansion and initial power spectrum amplitude, assembly history, and, where relevant, the WDM cutoff scale in the initial power spectrum. From this equation, one can explain the dependence of individual-halo concentration on redshift, halo mass, the initial power spectrum, cosmology, and halo MAH. The dependencies on redshift and initial power-spectrum are similar to those presented in Paper I: halos at higher redshift exhibit smaller concentrations because D(z)D(z) decreases with increasing zz, while halos in cosmologies with a larger initial power-spectrum normalization σ8\sigma_8 (and hence larger σ(M)\sigma(M)) yield larger concentrations. Meanwhile, more-massive halos have smaller concentrations, given that σ(M)\sigma(M) decreases with MM. However, unlike the formulae for peak-halo concentration presented in Paper I, the equation above quantitatively captures the relation with individual halo MAH.

As expected, a halo formed at an earlier epoch (with larger zfz_f and hence larger ωf\omega_f) possesses a larger concentration. The above dependencies can be understood theoretically in such a way that a higher σ8\sigma_8 and an earlier individual formation epoch zfz_f both imply that the halo’s inner core collapsed at an earlier cosmic time when the background universe was denser, imprinting a more compact scale radius rsr_s and hence a higher concentration. Conversely, toward lower redshift zz, the drop in cosmic background density coupled with late-time envelope growth expands the outer radius rΔr_\Delta relative to the stabilized inner core, systematically increasing cΔ=rΔ/rsc_\Delta= r_\Delta/r_s. Dependencies on other cosmological parameters are more complex, as they are embedded within the relations for D(z)D(z) and σ(M)\sigma(M).

While Figure 1 demonstrates the overall tightness of the global regression across the pooled dataset, combining all simulations and epochs into a single plot can obscure potential systematic biases within specific cosmological models or individual redshifts. To verify that Equation (5) holds unbiasedly across distinct regimes without recalibration, we apply the globally calibrated relation (with the fixed coefficients in Table 2) to individual relaxed halos in the TNG (ΛCDM) and ΛWDM simulations at separate redshifts. Figure 2 presents the predicted versus the measured concentrations for these two simulation suites.

Comparison of predicted and measured concentrations for relaxed halos in TNG and ΛWDM simulations

Figure 2. Comparison of the predicted and measured concentrations for relaxed halos in the TNG (Λ\LambdaCDM, left) and Λ\LambdaWDM (right) simulations. Different colors represent different redshifts. For each redshift, the outer and inner contours enclose 75% and 50% of the halos, respectively. The grey diagonal line marks the one-to-one relation. Across both cosmologies, the scatter around equality remains remarkably tight at σ≈0.09\sigma\approx0.09 dex.

For both cosmologies, halos at different redshifts follow a tight sequence close to the one-to-one line almost without systematic tilts. The colored contours enclose 75% and 50% of the halos at each redshift, showing that the bulk of the population is accurately reproduced by the model. The typical scatter about the one-to-one relation remains remarkably stable at σ≲0.1\sigma\lesssim0.1 dex for both ΛCDM and ΛWDM across nearly all tested redshifts, apart from the cases at z=5z = 5.

The tight agreement in the ΛWDM case is especially noteworthy. Although the suppression of small-scale power substantially alters halo accretion histories and lowers the concentrations of low-mass halos, our cutoff parameter AhmA_{\mathrm{hm}}, with aa fixed to the calibrated peak-model value, successfully absorbs this suppression. This indicates that individual halo concentration variations across cosmologies can be reliably modeled by combining linear variance, scaled formation epoch, and cutoff scale. With the individual relation verified, we now turn to generating the halo concentration distributions via Monte Carlo sampling.

Predicting concentration distributions from formation histories

The individual halo relation in Equation (5) can be utilized not only to predict the concentration of a given halo, but also to reconstruct the concentration distribution at fixed mass and redshift. If the scatter in concentration is predominantly driven by halo assembly history, then a distribution of formation variables, P(ωf)P(\omega_{\mathrm{f}}), should map directly into the concentration distribution via Monte Carlo realizations.

To inspect the input distribution, Figure 3 compares the probability density functions (PDFs) of ωf\omega_{\mathrm{f}} measured directly from cosmological simulations across redshifts (z=0z = 0–5) with the theoretical expectation from standard EPS theory at z=0z = 0 (dashed black curves). Two key physical insights emerge from this comparison.

Normalized probability density functions of the scaled formation variable

Figure 3. Normalized probability density functions (PDFs) of the scaled formation variable ωf\omega_{\mathrm{f}} from z=0z = 0 to z=5z = 5. The left and right panels show the numerical simulation measurements (colored histograms and curves) for Λ\LambdaCDM and Λ\LambdaWDM, respectively. These are compared with the theoretical expectation from standard EPS theory at z=0z = 0 (dashed black curves). The global fits across all redshifts using the generalized gamma distribution (Equation (6)) are displayed as red dashed curves. Notably, the simulated distributions peak at significantly higher values than standard EPS (highlighting a delayed-formation bias in EPS theory), and the WDM distribution is markedly narrower due to the suppression of small-scale perturbations.

First, standard EPS systematically predicts later halo formation times, peaking at ωf,peakEPS≈0.75\omega_{\mathrm{f,peak}}^{\mathrm{EPS}} \approx0.75 (as derived in the Appendix of Paper I), significantly lower than the actual peaks measured in the simulations (see Figure 3). Because ωf\omega_{\mathrm{f}} is defined such that larger values correspond to earlier collapse epochs (Equation (3)), this systematic deficit directly explains why EPS theory underestimates halo formation times relative to NN-body simulations (e.g., [23]). Second, the ωf\omega_{\mathrm{f}} distribution in WDM is markedly narrower than in Λ\LambdaCDM. This occurs because the primordial power spectrum in WDM is

severely suppressed below the half-mode scale MhmM_{\rm hm} due to dark matter thermal free-streaming. In the absence of small-scale perturbations, hierarchical, stochastic mergers of tiny sub-components are extinguished; instead, halos near the truncation scale form delayed, rapidly, and monolithically (e.g., P. Bode et al. [5]; A. Schneider et al. [37]; A. D. Ludlow et al. [24]). Because their non-linear collapse is tightly coupled to the epoch when the cutoff scale enters the non-linear regime, halo assembly histories in WDM become remarkably homogeneous, compressing the distribution into a substantially narrower width.

A closer inspection of Figure 3 indicates minor redshift variations: with increasing redshift, the simulated distributions shift slightly toward lower ωf\omega_{\rm f} and become marginally narrower, reflecting younger dynamical states and shorter available accretion timescales at early epochs. Nonetheless, the distributions maintain broad cosmological self-similarity across cosmic time. By definition (Equation (3)), the scaled formation variable is non-negative (ωf≥0\omega_{\rm f} \ge0), naturally vanishing at ωf=0\omega_{\rm f} = 0. We therefore parameterize the overall distribution of ωf\omega_{\rm f} across all redshifts using a generalized gamma distribution (E. W. Stacy [40]; P. Bett et al. [4]):

P(ωf)=C0(ωfω0)kexp⁡[−kp(ωfω0)p],(ωf≥0)(6)P(\omega_{\rm f}) = C_0 \left(\frac{\omega_{\rm f}}{\omega_0}\right)^k \exp\left[-\frac{k}{p}\left(\frac{\omega_{\rm f}}{\omega_0}\right)^p\right], \qquad(\omega_{\rm f} \ge0) \tag*{(6)}

where ω0≡ωf,peak\omega_0 \equiv\omega_{\rm f,peak} directly corresponds to the peak of the distribution, kk governs the low-ωf\omega_{\rm f} power-law rise, pp controls the exponential cutoff, and the normalization constant is given by

C0=pω0(k/p)(k+1)/pΓ(k+1p)(7)C_0 = \frac{p}{\omega_0}\frac{(k/p)^{(k+1)/p}}{\Gamma\left(\frac{k+1}{p}\right)} \tag*{(7)}

where Γ(x)=∫0∞tx−1e−t dt\Gamma(x) = \int_0^\infty t^{x-1}e^{-t}\,dt is the standard gamma function. Fitting Equation (6) across all redshifts yields (ω0,k,p)=(1.25,2.14,2.04)(\omega_0,k,p) = (1.25, 2.14, 2.04) for Λ\LambdaCDM and (1.21,1.82,2.70)(1.21, 1.82, 2.70) for Λ\LambdaWDM (shown as red dashed curves in Figure 3), corresponding to simulated peak values of ωf,peak≈1.25\omega_{\rm f,peak} \approx1.25 and ≈1.21\approx1.21, respectively. For simplicity, we adopt this global fit across cosmic epochs, as the minor redshift dependencies have negligible impact on the reconstructed concentration statistics.

Equation (6) shares deep physical asymptotes with the theoretical EPS distribution, p(ωf)=2ωferfc⁡(ωf/2)p(\omega_{\rm f}) = 2\omega_{\rm f}\operatorname{erfc}(\omega_{\rm f}/\sqrt{2}) ([21 Paper I]). In the rarepeak, early-formation limit (ωf≫1\omega_{\rm f} \gg1), the complementary error function asymptotically expands as erfc⁡(x)∼e−x2/(πx)\operatorname{erfc}(x) \sim e^{-x^2}/(\sqrt{\pi}x), yielding an exponential tail p(ωf)∝exp⁡(−ωf2/2)p(\omega_{\rm f}) \propto\exp(-\omega_{\rm f}^2/2) governed by the Gaussian statistics of the linear density field. The exponential tail in Equation (6) faithfully preserves this physical asymptote, with the fitted exponent p≈2.04p \approx2.04 in Λ\LambdaCDM closely recovering the theoretical Gaussian decay p=2p = 2. In the low-ωf\omega_{\rm f} regime (ωf→0\omega_{\rm f} \to0), while standard EPS features a linear rise (p(ωf)∝ωfp(\omega_{\rm f}) \propto\omega_{\rm f}) driving a steep macroscopic slope toward its early theoretical peak (ωf,peakEPS≈0.75\omega_{\rm f,peak}^{\rm EPS} \approx0.75), the simulation distributions exhibit a quadratic suppression (k≈1.82k \approx1.82–2.142.14) and a broader rise toward a delayed peak. In addition, the steeper exponential tail in WDM (p≈2.70p \approx2.70) cleanly captures the enhanced cutoff suppression and the synchronized, rapid collapse of structures near the free-streaming scale.

We then perform Monte Carlo realizations to reproduce the concentration distribution from the formation history PDF P(ωf)P(\omega_{\rm f}). For each halo mass and redshift (which fix the proxies D(z)D(z) and σ(M)\sigma(M), as well as AhmA_{\rm hm}), random values of ωf\omega_{\rm f} are drawn from the PDF P(ωf)P(\omega_{\rm f}) and substituted into the empirical regression relation (Equation (5)) to synthesize individual halo concentrations and predict the c(M)c(M) distribution.

The upper panels of Figure 4 show the resulting Monte Carlo predictions when drawing random ωf\omega_{\rm f} from the theoretical EPS distribution for TNG at z=0z = 0–22. While the EPS-based model captures the overall qualitative trends, it systematically underpredicts halo concentrations, especially at the low-mass end. This offset arises directly from the underestimation of ωf\omega_{\rm f} by EPS theory.

Monte Carlo predictions of concentration–mass distributions for relaxed and all halos at redshifts $z=0$, $1$, and $2$

Figure 4. Monte Carlo predictions of concentration–mass distributions at z=0z = 0–22, generated by substituting random ωf\omega_{\mathrm{f}} draws into Equation (5). Left and right columns show relaxed and all halos, respectively. Rows indicate the ωf\omega_{\mathrm{f}} input source: theoretical EPS for Λ\LambdaCDM (top), simulation PDF for Λ\LambdaCDM (middle), and simulation PDF for Λ\LambdaWDM (bottom). Solid red curves display the peak concentration model (from Paper I for all halos; Appendix § 7 for relaxed halos), and red circles mark the extracted binned lognormal peak values. The color scale denotes ωf\omega_{\mathrm{f}}.

We then repeat this Monte Carlo reproduction using random ωf\omega_{\rm f} drawn directly from the simulation-measured PDF (Equation (6)). The middle and lower panels of Figure 4 show that these simulation-based Monte Carlo predictions eliminate this offset and track the peak concentration–mass relation of Paper I (red lines, corresponding to Paper I for all halos and our Appendix § 7 calibration for relaxed halos) and the extracted binned lognormal peak values (red circles) remarkably well across all masses and redshifts in both Λ\LambdaCDM and Λ\LambdaWDM. In WDM cosmologies (bottom row), incorporating simulation-drawn ωf\omega_{\rm f} along with AhmA_{\rm hm} accurately reproduces the flattening and turnover in halo concentrations below MhmM_{\rm hm}.

Comparison with other models

To assess the performance and applicability of our empirical formulation, we compare our individual halo predictions with previous concentration models, most notably the physically motivated model developed by A. D. Ludlow et al. [24]. The Ludlow et al. model relates the internal characteristic density of a dark matter halo to the background cosmic critical density at the collapse redshift z−2z_{-2}, defined as the epoch when the cumulative collapsed mass history (CMH) of the main progen-ેણ一级特黄大片 itor first assembled the core mass M−2M_{-2}. In their work, halo density profiles were modeled using an Einasto (J. Einasto [13]) profile with a fixed shape parameter of α=0.18\alpha= 0.18, and concentrations were defined relative to the cosmic critical density as c200c=r200c/r−2c_{200\mathrm{c}} = r_{200\mathrm{c}}/r_{-2}.

To ensure a rigorous and direct comparison, we recalibrate our empirical multi-linear relation (Equation (5)) using the exact same profile definition: concentrations derived by fitting an Einasto profile with fixed α=0.18\alpha= 0.18 within r200cr_{200\mathrm{c}}. Furthermore, because A. D. Ludlow et al. [24] evaluated linear density fluctuations using a real-space top-hat filter for all cosmologies (including WDM), in this subsection we also evaluate σ(M)\sigma(M) using the real-space top-hat filter to maintain an identical baseline. Because full dark matter particle snapshots across all redshifts are not publicly available for the MultiDark-Planck and Uchuu suites, we restrict this Einasto profile recalibration to relaxed halos from the TNG and our non-Λ\LambdaCDM simulations. The resulting Einasto-based empirical relation is given by

log⁡c200cEinasto≈0.46+0.89log⁡D(z)+0.63log⁡σ(M)+0.71log⁡(1+ωf)−0.44log⁡Ahm(8)\log c^{\mathrm{Einasto}}_{200\mathrm{c}} \approx0.46 + 0.89\log D(z) + 0.63\log\sigma(M) + 0.71\log(1+\omega_{\mathrm{f}}) - 0.44\log A_{\mathrm{hm}} \tag*{(8)}

with the WDM cutoff parameter fixed to a=5.59a = 5.59 as determined from the relaxed peak model under the 200c200\mathrm{c} boundary definition. Figure 5 shows the regression performance of Equation (8) across the TNG and non-Λ\LambdaCDM cosmological simulation datasets. The predicted concentrations tightly follow the measured Einasto values across the entire range z=0z = 0–55, exhibiting an exceptionally small residual scatter that increases from σlog⁡c≈0.06\sigma_{\log c} \approx0.06 dex at z=0z = 0 to ≈0.13\approx0.13 dex at z=5z = 5. This confirms that our four-variable formulation accurately captures halo concentrations regardless of whether an NFW or an Einasto density profile is adopted.

Global fit of the empirical concentration model for relaxed halos

Figure 5. Global fit of our empirical concentration model for relaxed halos under the Δh=200c\Delta_{\mathrm{h}} = 200\mathrm{c} definition, calibrated using concentrations derived from an Einasto profile with α=0.18\alpha= 0.18. The distribution combines individual halos from the TNG and non-Λ\LambdaCDM simulations from z=5z = 5 to z=0z = 0, visualized via hexagonal binning color-coded by the mean redshift in each cell. The dashed line marks the one-to-one diagonal, and contours enclose 95% and 68% of halos. The lower sub-panel shows the binned residual scatter, achieving an average dispersion of σlog⁡c≈0.098\sigma_{\log c} \approx0.098 dex.

In A. D. Ludlow et al. [24] (their Figure 10), the authors evaluated their model against numerical simulations (the COCO-COLD and COCO-WARM suites) only across three low-to-intermediate redshifts: z=0z = 0–22. Over this redshift baseline (z≤2z \le2), their model was shown to reproduce individual halo concentrations with an rms scatter of σ≲0.09\sigma\lesssim0.09 dex. However, how their model behaves at higher redshifts (z>2z > 2) was not presented in their original paper. Here, we extend the comparison between the two models across all redshifts (z=0z = 0–55) for both Λ\LambdaCDM (TNG) and Λ\LambdaWDM simulations, as presented in Figure 6.

Comparison of the 200c concentrations predicted by the Einasto-calibrated empirical model and the model of A. D. Ludlow et al. (2016) for relaxed halos

Figure 6. Comparison of the 200c concentrations predicted by our Einasto-calibrated empirical model (Equation (8)) and those predicted by the model of A. D. Ludlow et al. [24] for relaxed halos in the TNG (Λ\LambdaCDM, left) and Λ\LambdaWDM (right) simulations. Different colors represent redshifts from z=0z = 0 to z=5z = 5. Contours enclose 75% and 50% of the halo population, while the grey diagonal lines denote equality. In both cosmological frameworks, the two independent models demonstrate remarkable broad consistency across the entire dynamic range from z=0z = 0 to z=5z = 5.

As shown in Figure 6, the two independently developed models show broad and remarkable consistency across both cosmologies. Over the redshift interval z=0z = 0–22 where the Ludlow et al. model was originally calibrated, the predictions of both models are in excellent agreement, closely aligning along the one-to-one diagonal. At higher redshifts (z=3z = 3–55), both models continue to track the same overall sequence over several orders of magnitude in halo mass. In Λ\LambdaWDM (right panel), the two models remain remarkably consistent along equality all the way to z=5z = 5. In Λ\LambdaCDM (left panel), the contours exhibit a modest shift at z≥3z \ge3, where our model predicts slightly higher concentrations by ∼0.08\sim0.08–0.100.10 dex. This difference remains comparable to the typical halo-to-halo intrinsic scatter (∼0.15\sim0.15–0.200.20 dex) at these early epochs, confirming that both frameworks capture the primary physical drivers of halo internal structure.

While both approaches demonstrate that halo assembly history predominantly governs halo concentration, they represent complementary modeling philosophies. The Ludlow et al. model is grounded in a physical linkage between inner density and the critical density at the CMH crossing epoch, which requires tracing the complete hierarchy of progenitor merger branches and numerically solving a transcendental equation for every individual halo. In contrast, our empirical relation provides an explicit, closed-form algebraic formula that requires only four readily accessible parameters (D(z)D(z), σ(M)\sigma(M), ωf\omega_{\mathrm{f}}, and AhmA_{\mathrm{hm}}). This compact mathematical structure makes our model computationally lightweight and straightforward to integrate into large-scale semi-analytic galaxy formation models and cosmological mock catalogs.

We also compare our results with the machine-learning approach of T. Zhang et al. [44], who trained neural networks to predict halo concentrations from complete MAHs over z=0z = 0–22. While their neural network achieves an RMSE of 0.0845 at z=0z = 0, the scatter increases to 0.1287 at z=2z = 2, and predictions beyond redshift 2 were not provided. Our simple analytical relation achieves a competitive, nearly uniform scatter across z=0z = 0–55 with clear physical interpretability, providing a transparent and robust framework for individual halo concentration predictions across diverse cosmological scenarios.

Discussion and Conclusions

In this work, we have established a concise, empirical formulation for predicting the internal concentrations of individual dark matter halos across both cold and warm dark matter cosmologies spanning from redshift z=5z = 5 down to z=0z = 0. By explicitly decoupling cosmological expansion via the linear growth factor D(z)D(z), mass scale through the linear fluctuation amplitude σ(M)\sigma(M), individual mass accretion histories through the EPS-scaled formation variable ωf\omega_{\mathrm{f}}, and the small-scale damping scale via AhmA_{\mathrm{hm}}, our model accurately captures the halo-to-halo concentration scatter and its cosmological evolution.

Calibrated against an extensive suite of cosmological NN-body simulations, our model achieves a typical residual dispersion of ≈0.08\approx0.08–0.100.10 dex for relaxed halos across all examined redshifts. For the complete, unselected halo population, the scatter remains similarly narrow (≈0.09\approx0.09 dex) at z=0z = 0 but exhibits pronounced redshift evolution, increasing to ≈0.18\approx0.18 dex at z=5z = 5 due to the rising fraction of dynamically disturbed systems at early epochs. Both virial and 200c boundary definitions are well described by this framework. When applied to reconstruct concentration distributions, adopting ωf\omega_{\mathrm{f}} measured directly from simulation merger trees faithfully reproduces both the empirical distributions and the peak-concentration relations from our previous work. In contrast, standard EPS theory systematically underpredicts concentrations due to an inherent delayed-formation bias.

In summary, using D(z)D(z), σ(M)\sigma(M), and ωf\omega_{\mathrm{f}} as proxies for halo-identification redshift, halo mass, and halo assembly epoch, respectively, our model can accurately predict the concentration parameter for individual halos given their MAH. The dependencies on redshift, halo mass, cosmology, the initial power-spectrum and MAH are embedded within these three proxies. Compared with the model of A. D. Ludlow et al. [24], which requires tracking the full progenitor-merger trees of the target halo, our formulation provides comparable predictive fidelity via an explicit closed-form algebraic relation that is also computationally convenient and has been extended to redshift 5. This makes our framework particularly well-suited for rapid implementation within large-scale cosmological mock pipelines and semi-analytic galaxy-formation models. By coupling our closed-form relation with merger-tree formation epochs or Monte-Carlo draws from our calibrated P(ωf)P(\omega_f) distribution, galaxy-formation models and survey mock catalogs can self-consistently propagate halo assembly histories into internal density profiles, galaxy structural scaling relations, and predictions for small-scale clustering or lensing. Conversely, given observational measurements of halo mass and concentration parameters, for example, from lensing or dynamical analysis, our model enables us to infer the merger history of the target halo. A user-friendly code package implementing the concentration models of Paper I and this work is publicly available on GitHub.4

Although our empirical formulation accounts for the primary driver of concentration diversity and stabilizes the typical scatter to ≈0.08\approx0.08–0.100.10 dex for relaxed halos, a non-negligible fraction of the variance remains unexplained, since mass-accretion histories alone repro-

duce only 40%–50% of the total scatter within the CDM framework (c.f., A. D. Ludlow et al. [24]). Other plausible contributors to halo-structure variance include environmental modulations such as large-scale tidal fields and anisotropic shear. Our framework offers a natural pathway to investigate how large-scale environment modulates halo structure. Because halos in different cosmic web environments (e.g., nodes, sheets, filaments and voids) experience systematically different assembly histories—closely linked to halo assembly bias (e.g., L. Gao et al. [15])—our model can quantify how environment-dependent shifts in the conditional ωf\omega_f distribution drive concentration variations. Disentangling how the cosmic environment shapes halo internal profiles both through and beyond their mass accretion histories represents a promising direction for future work.

Acknowledgements

The authors thank the TNG, MultiDark-Planck, Uchuu projects and Dr. Peter Behroozi for publicly releasing the Rockstar and Consistent trees codes. This research has been supported by the NSFC grants (Nos. 12573007, 12073089). WSZ is supported by National SKA Program of China (grant Nos. 2025SKA0150104) and NSFC grant 12595314. Parts of the simulations and all of the analysis in this work were carried out at the Kunlun & Loong HPC facility of the School of Physics and Astronomy, Sun Yat-sen University.

Appendix

Calibration of the Peak Concentration Model for Relaxed Halos

In § 3.2 (and Figure 4), we test whether our individual halo concentration relation (Equation (5)) can statistically reconstruct the global concentration distribution when combined with the assembly history distribution P(ωf)P(\omega_f). To rigorously evaluate whether the predicted concentration sequences match the typical halo population across mass, redshift, and cosmology, an independent population-level benchmark is required. For this purpose, we benchmark our reconstructed distributions against the universal peak-concentration framework established in Paper I.

Furthermore, while Paper I calibrated the peak-concentration relation exclusively for the complete, unselected halo population (the all-halo sample), this work places primary emphasis on relaxed halos. Therefore, to provide the exact reference benchmark for relaxed halos in Figure 4 and to physically ground the WDM cutoff parameter aa in Equation (4), here we summarize the peak model formulation and report its newly calibrated parameters for relaxed halos.

Model Formulation

Following Paper I, the peak concentration of halos is characterized by their collapse history through the formation peak-height framework. The characteristic linear growth factor evaluated at the peak formation epoch, D(zf,peak)D(z_{\mathrm{f,peak}}), is decomposed into a baseline CDM component and a WDM suppression factor:

D(zf,peak)=D(zf,peak)CDM E(MΔ,Mhm),(9)D(z_{\mathrm{f,peak}}) = D(z_{\mathrm{f,peak}})_{\mathrm{CDM}}\,\mathcal{E}(M_{\Delta}, M_{\mathrm{hm}}), \tag*{(9)}

where the baseline CDM growth factor follows

D(zf,peak)CDM=D(z)[a1(MΔ1012h−1M⊙)b1+a2D(z)b2],(10)D(z_{\mathrm{f,peak}})_{\mathrm{CDM}} = D(z)\left[a_1\left(\frac{M_{\Delta}}{10^{12}h^{-1}M_{\odot}}\right)^{b_1} + a_2D(z)^{b_2}\right], \tag*{(10)}

and the WDM modification factor is parameterized as

E(MΔ,Mhm)=1.0+η(MhmMΔ)μ.(11)\mathcal{E}(M_{\Delta}, M_{\mathrm{hm}}) = 1.0 + \eta\left(\frac{M_{\mathrm{hm}}}{M_{\Delta}}\right)^{\mu}. \tag*{(11)}

Here MhmM_{\mathrm{hm}} is the half-mode mass of the WDM model (with Mhm=0M_{\mathrm{hm}} = 0 and E=1\mathcal{E} = 1 for cosmologies without a WDM-like cutoff).

The peak concentration cpeakc_{\mathrm{peak}} is then universally described as a power law of the effective formation peak height νeff\nu_{\mathrm{eff}}:

cpeak=Apνeff−αp+c0,(12)c_{\mathrm{peak}} = A_{\mathrm{p}}\nu_{\mathrm{eff}}^{-\alpha_{\mathrm{p}}} + c_0, \tag*{(12)}

where νeff\nu_{\mathrm{eff}} incorporates the small-scale cutoff suppression via

νeff=νpeakAhm,νpeak≡δsc(0)D(z)D(zf,peak)σ(M).(13)\nu_{\mathrm{eff}} = \nu_{\mathrm{peak}}A_{\mathrm{hm}}, \qquad\nu_{\mathrm{peak}} \equiv\frac{\delta_{\mathrm{sc}}(0)}{D(z)D(z_{\mathrm{f,peak}})\sigma(M)}. \tag*{(13)}

Calibrated Parameter Sets

For the complete all-halo sample, the parameters were established in Paper I using the sharp-kk filter for WDM. For relaxed halos, we apply the exact same calibration procedure to the relaxed halo catalogs across all our cosmological simulations from z=5z = 5 to z=0z = 0.

The resulting best-fitting parameter sets for both relaxed halos and the complete all-halo population, under both the virial (Δvir\Delta_{\mathrm{vir}}) and 200 times critical density (Δ200c\Delta_{200\mathrm{c}}) boundary definitions, are summarized in Table 3.

SampleBoundary Δh\Delta_{\mathrm{h}}Baseline D(zf,peak)CDMD(z_{\mathrm{f,peak}})_{\mathrm{CDM}}WDM Factor E\mathcal{E}Peak Relation cpeak(νeff)c_{\mathrm{peak}}(\nu_{\mathrm{eff}})Cutoff
Relaxed halosΔvir\Delta_{\mathrm{vir}}0.210.100.33−0.48-0.480.141.0713.200.932.104.77
All halosΔvir\Delta_{\mathrm{vir}}0.200.120.36−0.43-0.430.121.0912.031.092.543.98
Relaxed halosΔ200c\Delta_{200\mathrm{c}}0.200.110.33−0.50-0.500.151.049.970.681.645.59
All halosΔ200c\Delta_{200\mathrm{c}}0.200.110.35−0.47-0.470.131.068.810.952.313.06

Table 3. Best-fitting parameters of the formation peak-height concentration model across two halo definitions (Δvir\Delta_{\mathrm{vir}} and Δ200c\Delta_{200\mathrm{c}}). Parameters for the all-halo sample are taken from Paper I, while those for relaxed halos are newly calibrated in this work. Baseline D(zf,peak)CDMD(z_{\mathrm{f,peak}})_{\mathrm{CDM}}: a1a_1, b1b_1, a2a_2, b2b_2; WDM Factor E\mathcal{E}: η\eta, μ\mu; Peak Relation cpeak(νeff)c_{\mathrm{peak}}(\nu_{\mathrm{eff}}): ApA_{\mathrm{p}}, αp\alpha_{\mathrm{p}}, c0c_0; Cutoff: aa.

Note. The baseline CDM growth factor is given by Equation (10), the WDM modification factor E\mathcal{E} by Equation (11), and the peak-concentration relation by Equation (12).

As shown in Table 3, relaxed halos consistently exhibit a higher normalization ApA_{\mathrm{p}} and a flatter slope αp\alpha_{\mathrm{p}} compared to the all-halo population, reflecting their systematically higher concentrations and earlier collapse epochs. Crucially, the cutoff coefficients aa determined here (last column of Table 3) directly set the scaling of AhmA_{\mathrm{hm}} in Equation (4), ensuring seamless physical consistency between our individual halo model and the population peak statistics.

References

  1. [1]Anbajagane, D., Evrard, A. E., & Farahi, A. 2022, MNRAS, 509, 3441, doi: 10.1093/mnras/stab3177DOI
  2. [2]Behroozi, P. S., Wechsler, R. H., & Wu, H.-Y. 2013a, ApJ, 762, 109, doi: 10.1088/0004-637X/762/2/109DOI
  3. [3]Behroozi, P. S., Wechsler, R. H., Wu, H.-Y., et al. 2013b, ApJ, 763, 18, doi: 10.1088/0004-637X/763/1/18DOI
  4. [4]Bett, P., Eke, V., Frenk, C. S., et al. 2007, MNRAS, 376, 215, doi: 10.1111/j.1365-2966.2007.11432.xDOI
  5. [5]Bode, P., Ostriker, J. P., & Turok, N. 2001, ApJ, 556, 93, doi: 10.1086/321541DOI
  6. [6]Bryan, G. L., & Norman, M. L. 1998, ApJ, 495, 80, doi: 10.1086/305262DOI
  7. [7]Bullock, J. S., Kolatt, T. S., Sigad, Y., et al. 2001, MNRAS, 321, 559, doi: 10.1046/j.1365-8711.2001.04068.xDOI
  8. [8]Child, H. L., Habib, S., Heitmann, K., et al. 2018, ApJ, 859, 55, doi: 10.3847/1538-4357/aabf95DOI
  9. [9]Diemer, B., & Joyce, M. 2019, ApJ, 871, 168, doi: 10.3847/1538-4357/aafad6DOI
  10. [10]Diemer, B., & Kravtsov, A. V. 2015, ApJ, 799, 108, doi: 10.1088/0004-637X/799/1/108DOI
  11. [11]Duffy, A. R., Schaye, J., Kay, S. T., & Dalla Vecchia, C. 2008, MNRAS, 390, L64, doi: 10.1111/j.1745-3933.2008.00537.xDOI
  12. [12]Dutton, A. A., & Macciò, A. V. 2014, MNRAS, 441, 3359, doi: 10.1093/mnras/stu742DOI
  13. [13]Einasto, J. 1965, Trudy Astrofizicheskogo Instituta Alma-Ata, 5, 87
  14. [14]Eke, V. R., Navarro, J. F., & Steinmetz, M. 2001, ApJ, 554, 114, doi: 10.1086/321345DOI
  15. [15]Gao, L., Navarro, J. F., Cole, S., et al. 2008, MNRAS, 387, 536, doi: 10.1111/j.1365-2966.2008.13277.xDOI
  16. [16]Heitmann, K., Higdon, D., White, M., et al. 2009, ApJ, 705, 156, doi: 10.1088/0004-637X/705/1/156DOI
  17. [17]Heitmann, K., White, M., Wagner, C., Habib, S., & Higdon, D. 2010, ApJ, 715, 104, doi: 10.1088/0004-637X/715/1/104DOI
  18. [18]Ishiyama, T., Prada, F., Klypin, A. A., et al. 2021, MNRAS, 506, 2410, doi: 10.1093/mnras/stab1755DOI
  19. [19]Jing, Y. P. 2000, ApJ, 535, 30, doi: 10.1086/308809DOI
  20. [20]Klypin, A., Yepes, G., Gottlöber, S., Prada, F., & Heß, S. 2016, MNRAS, 457, 4340, doi: 10.1093/mnras/stw248DOI
  21. [21]Lacey, C., & Cole, S. 1993, MNRAS, 262, 627, doi: 10.1093/mnras/262.3.627DOI
  22. [22]Lewis, A., Challinor, A., & Lasenby, A. 2000, ApJ, 538, 473, doi: 10.1086/308179
  23. [23]Lin, W. P., Jing, Y. P., & Lin, L. 2003, MNRAS, 344, 1327, doi: 10.1046/j.1365-8711.2003.06924.xDOI
  24. [24]Ludlow, A. D., Bose, S., Angulo, R. E., et al. 2016, MNRAS, 460, 1214, doi: 10.1093/mnras/stw1046DOI
  25. [25]Ludlow, A. D., Navarro, J. F., Angulo, R. E., et al. 2014, MNRAS, 441, 378, doi: 10.1093/mnras/stu483DOI
  26. [26]Ludlow, A. D., Navarro, J. F., Li, M., et al. 2012, MNRAS, 427, 1322, doi: 10.1111/j.1365-2966.2012.21892.xDOI
  27. [27]Ludlow, A. D., Schaye, J., & Bower, R. 2019, MNRAS, 488, 3663, doi: 10.1093/mnras/stz1821DOI
  28. [28]Macciò, A. V., Dutton, A. A., van den Bosch, F. C., et al. 2007, MNRAS, 378, 55, doi: 10.1111/j.1365-2966.2007.11720.xDOI
  29. [29]Marinacci, F., Vogelsberger, M., Pakmor, R., et al. 2018, MNRAS, 480, 5113, doi: 10.1093/mnras/sty2206DOI
  30. [30]Naiman, J. P., Pillepich, A., Springel, V., et al. 2018, MNRAS, 477, 1206, doi: 10.1093/mnras/sty618DOI
  31. [31]Navarro, J. F., Frenk, C. S., & White, S. D. M. 1996, ApJ, 462, 563, doi: 10.1086/177173DOI
  32. [32]Navarro, J. F., Frenk, C. S., & White, S. D. M. 1997, ApJ, 490, 493, doi: 10.1086/304888DOI
  33. [33]Nelson, D., Pillepich, A., Springel, V., et al. 2018, MNRAS, 475, 624, doi: 10.1093/mnras/stx3040DOI
  34. [34]Neto, A. F., Gao, L., Bett, P., et al. 2007, MNRAS, 381, 1450, doi: 10.1111/j.1365-2966.2007.12381.xDOI
  35. [35]Pillepich, A., Nelson, D., Hernquist, L., et al. 2018, MNRAS, 475, 648, doi: 10.1093/mnras/stx3112DOI
  36. [36]Reed, D., Governato, F., Verde, L., et al. 2005, MNRAS, 357, 82, doi: 10.1111/j.1365-2966.2005.08612.xDOI
  37. [37]Schneider, A., Smith, R. E., Macciò, A. V., & Moore, B. 2012, MNRAS, 424, 684, doi: 10.1111/j.1365-2966.2012.21252.xDOI
  38. [38]Schneider, A., Smith, R. E., & Reed, D. 2013, MNRAS, 433, 1573, doi: 10.1093/mnras/stt829DOI
  39. [39]Springel, V., Pakmor, R., Pillepich, A., et al. 2018, MNRAS, 475, 676, doi: 10.1093/mnras/stx3304DOI
  40. [40]Stacy, E. W. 1962, The Annals of mathematical statistics, 1187
  41. [41]Wang, D.-Z., Lin, W., & Luan, T.-C. 2026, ApJ, 1006, 61, doi: 10.3847/1538-4357/ae81a2DOI
  42. [42]Wang, K., Mo, H. J., Chen, Y., & Schave, J. 2024, MNRAS, 527, 10760, doi: 10.1093/mnras/stad3927DOI
  43. [43]Wechsler, R. H., Bullock, J. S., Primack, J. R., Kravtsov, A. V., & Dekel, A. 2002, ApJ, 568, 52, doi: 10.1086/338765
  44. [44]Zhang, T., Mao, T., Xu, W., & Li, G. 2025, Universe, 11, 37, doi: 10.3390/universe11020037DOI
  45. [45]Zhao, D. H., Jing, Y. P., Mo, H. J., & Börner, G. 2003a, ApJL, 597, L9, doi: 10.1086/379734DOI
  46. [46]Zhao, D. H., Jing, Y. P., Mo, H. J., & Börner, G. 2009, ApJ, 707, 354, doi: 10.1088/0004-637X/707/1/354DOI
  47. [47]Zhao, D. H., Mo, H. J., Jing, Y. P., & Börner, G. 2003b, MNRAS, 339, 12, doi: 10.1046/j.1365-8711.2003.06135.xDOI

Paper details

vlnc:W_7jz7rpwz
Version
v1 · latest
Published
Fresh — added to the Hub within the last 30 days
Added
License
CC BY 4.0
Usage

views · 0

downloads · 0

A view counts once per browser per day, a download each time the PDF is requested.

  • Physics and Astronomy

Not recorded for this paper

View statement
Looking for related papers…
Contents