betaFeedback

WorksPhysics and Astronomy

Effect of galaxy-informed merger timescales on massive black hole binaries in LISA

Abstract

Over a decade of direct N-body simulations tell us that massive black hole binary (MBHB) merger timescales are strongly dependent on the structural and kinematic properties of the host galaxy merger remnant. The physical stages of stellar scattering and gravitational wave (GW) emission of MBH coalescence occur below the resolution limit of most cosmological simulations, requiring post-processing to achieve accurate merging times. We apply a recently developed galaxy-informed merger timescale for MBHBs to the Romulus25 simulation and compare with current MBH merger rate and event rate estimates for the upcoming Laser Interferometer Space Antenna (LISA). We find a median stellar scattering time of ~ 10^8.2 yr^-1, roughly an order of magnitude faster than those produced by timescales that rely on extrapolations of the stellar density profile. We predict a merger rate of ~ 11.5 yr^-1 corresponding to 46 merger events over a 4-year LISA mission. We also investigate the importance of the relative orbital inclination of the MBHB with respect to the host galaxy and of nuclear star cluster (NSC) modeling, finding that the quickest merger events correspond to MBHBs that counterrotate and galaxies assigned to host NSCs, which may be important for host galaxy identification in the multi-messenger era.

Introduction

Galaxies can grow by merging together, which offers opportunities in turn to form massive black hole binaries (MBHBs) (Begelman et al. 1980). The journey of the two MBHs toward pairing and coalescence is typically categorized into phases depending on the mechanism dominating the energy loss of the binary. In the canonical picture for MBHBs (104M⊙≲Mtot≲107M⊙10^{4} M_{\odot} \lesssim M_{\mathrm{tot}} \lesssim10^{7} M_{\odot}), dynamical friction dominates on ∼kpc\sim\mathrm{kpc} scales, driving the MBHs towards the center of their newly-shared host (Chandrasekhar 1943; Binney & Tremaine 1987). Stellar scattering effects become important on ∼10\sim10–11 pc scales, hardening the binary further (provided there is a sufficient supply of stars) (Milosavljević & Merritt 2003). The presence of an accretion disk on ∼\simsub-pc scales in gas-rich nuclei may also aid the hardening of the binary (Mayer et al. 2007; Cuadra et al. 2009; Haiman et al. 2009; Siwek et al. 2023). Once ∼\simpc separations are reached, the decay is dominated by the emission of gravitational waves (GWs) until coalescence (Peters & Mathews 1963).

The timescale governing each of these phases is complicated, however, and is highly dependent on the properties of the host galaxy. Dynamical friction is enhanced when MBHs are encased in a dense nuclear star cluster (NSC), increasing the effective mass and invoking tidal distortion that sinks the MBHs quickly (Ogiya et al. 2020; Mukherjee et al. 2025). Spikes from self-interacting dark matter may also aid the efficiency of sinking (Alonso-Álvarez et al. 2024). In contrast, MBHs can fail to sink for a multitude of reasons, including (but not limited to) getting trapped in galactic bar-induced resonances (Bortolas et al. 2022; Herrero et al. 2025), encountering star-forming clumps in non-relaxed systems, as would be common at high redshift (Ma et al. 2021), or being stripped of their associated stellar systems in minor mergers, leaving them “wandering” for longer than the age of the Universe (Bellovary et al. 2010; Zhou et al. 2025).

At scales where the hardening is dominated by stellar scattering, NN-body simulations and analytic estimates alike show that non-spherical galactic potentials are required to overcome the “final parsec problem” [129] (Berczik et al. 2006; Khan et al. 2013; Li et al. 2015; Gualandris et al. 2017). Additionally, any degree of rotation of the stellar component (Holley-Bockelmann & Khan 2015; Varisco et al.

© 2026 The Authors 2021), inclination of the MBHB with respect to the host (Mirza et al. 2017), and the presence of a nuclear star cluster (NSC) (Khan & Holley-Bockelmann 2021; Khan et al. 2026) also encourage efficient hardening. When embedded in a gaseous accretion disk, the orbital evolution of the binary depends on both binary properties, like mass ratio and initial eccentricity (Siwek et al. 2023), as well as the disk structure and thermodynamic properties (Tiede et al. 2020; D’Orazio & Duffell 2021; Dittmann et al. 2026; Clyburn & Zrake 2026). Simulations in the thin disk regime (which is expected for astrophysical disks surrounding MBHs in gas rich nuclei) do not yet agree on the sign of the imparted torque on the binary within the full range of possible parameter space (Duffell et al. 2024; Derdzinski et al. 2025). All of these effects combine in highly nonlinear ways, requiring expensive detailed modeling to achieve self-consistent predictions, albeit only for a few galaxy types at a time.

With their rich variety of host galaxy properties and ancestries, cosmological simulations offer a promising way to investigate the broader possible range of MBH coalescence times. A large challenge with these models, however, is that many of the aforementioned physical processes occur below the resolution limit of the simulation, which can be as large as ∼\sim few kpc. Instead, to prolong the lifetime of the MBHB, simulations either invoke sub-grid models, which operate on-the-fly, or post-processing models, which can be performed on simulation outputs. Both of these methods require analytical estimates for MBH merger timescales since the dynamics are not resolved outright.

Sesana & Khan (2015) (hereafter SK15) propose one such recipe for the merger timescale. They conjecture that the hardening parameter, (SS), defined as the rate of change of the inverse of the binary semi-major axis (aa), is equivalent that for a MBHB in an unbound, homogeneous stellar system (Quinlan 1996). The relation follows

SSK15=ddt(1a)=HGρinfσinf,(1)S_{\mathrm{SK15}} = \frac{d}{dt}\left(\frac{1}{a}\right) = H\frac{G\rho_{\mathrm{inf}}}{\sigma_{\mathrm{inf}}}, \tag*{(1)}

where H≈15H \approx15–2020 is a dimensionless rate, GG is the gravitational constant, and ρinf\rho_{\mathrm{inf}} and σinf\sigma_{\mathrm{inf}} are the stellar density and velocity dispersion measured at the binary influence radius (rinfr_{\mathrm{inf}}), which is defined as the radius surrounding the binary that encompasses twice the binary’s mass in stars.

This timescale has been well-used in cosmological simulations (e.g. Kelley et al. 2017b) and semi-analytic models (e.g. Biava et al. 2019; Bortolas et al. 2021) alike, as one can estimate the coalescence time using properties easily extractable from the stellar density profile of the host. However, the resolution limit in cosmological simulations is oftentimes orders of magnitude larger than rinfr_{\mathrm{inf}} of the binary. This requires us to either use analytic prescriptions or scaling relations for ρinf\rho_{\mathrm{inf}} and σinf\sigma_{\mathrm{inf}} or to inwardly extrapolate galaxy properties over very large scales.

Another recipe by Vasiliev et al. (2015) (hereafter V15) considers a separation-dependent hardening parameter that follows

SV15(a)=μSinf(a/ah)ν,(2)S_{\mathrm{V15}}(a) = \mu S_{\mathrm{inf}}(a/a_{\mathrm{h}})^{\nu}, \tag*{(2)}

where μ\mu is the filling fraction of the loss cone (fiducial value μ=0.3\mu= 0.3) and ν\nu determines the decay of SS with aa as the loss-cone depletes (fiducial value ν=0.4\nu= 0.4). SinfS_{\mathrm{inf}} is the full-loss-cone hardening rate at rinfr_{\mathrm{inf}}, obtained by substituting σinf2∼GMtot/rinf\sigma_{\mathrm{inf}}^2 \sim GM_{\mathrm{tot}}/r_{\mathrm{inf}} and ρinf=2Mtot/rinf3\rho_{\mathrm{inf}} = 2M_{\mathrm{tot}}/r_{\mathrm{inf}}^3 into Eq. (1):

Sinf=A(GMtot)1/2rinf−5/2(3)S_{\mathrm{inf}} = A(GM_{\mathrm{tot}})^{1/2}r_{\mathrm{inf}}^{-5/2} \tag*{(3)}

with a fiducial A=4A = 4. Lastly, aha_{\mathrm{h}} is the hard binary separation, i.e. the separation at which the binding energy of the binary increases rather than decreases with stellar interactions. aha_{\mathrm{h}} depends on the binary mass ratio (q=M2/M1≤1q = M_2/M_1 \leq1), and is given by ah=q4(1+q)2rinfa_{\mathrm{h}} = \frac{q}{4(1+q)^2}r_{\mathrm{inf}}.

The V15 timescale has been used in some form for post-processing delays in cosmological simulations with orbital information (e.g. [33] (Katz et al. 2020; Chen et al. 2022b)). Despite the control that this method has over the effect of loss-cone depletion in slowing the scattering process, it cannot reproduce the conditions of a rotating stellar background that are thought to bring the binary closer to the full loss-cone regime again (Holley-Bockelmann & Khan 2015; Varisco et al. 2021). Similarly to SK15, it also requires either analytic prescriptions or inwards extrapolation to measure rinfr_{\mathrm{inf}} and σinf\sigma_{\mathrm{inf}}.

In this paper, we adopt a galaxy-informed timescale proposed in Holley-Bockelmann et al. (2025) (hereafter HB25) which incorporates kinematic and structural properties of the host galaxy and orbital information about the binary into the estimate for the merger timescale through the late dynamical friction and stellar scattering phases of MBH coalescence. By relying largely on bulk host galaxy properties that are ostensibly simple to obtain from simulations and observations alike, the timescale aims to incorporate the galaxy properties relevant to stellar scattering while maintaining relative simplicity of implementation. It is important then to test the timescale on a cosmologically-realized population of host galaxies with realistic stellar structures and kinematics. Our choice of simulation, Romulus25 (Tremmel et al. 2017) also succeeds in accurately tracing MBH dynamics down to sub-kpc scales (Tremmel et al. 2015), offering a diverse array of MBHB orbital configurations as well (see Figure 1).

Representative orbit of a MBHB around a Milky Way mass galaxy in Romulus25

Figure 1. Representative orbit of a MBHB around a Milky Way mass galaxy in Romulus25. The left panel shows the face-on view with respect to the galaxy and the right panel shows the side-on view, oriented such that the stellar angular momentum vector points in the positive zz direction. The primary has a mass of M1∼3.3×108M⊙M_1 \sim3.3 \times10^{8}M_\odot and its trajectory is shown by the thick line; the secondary has a mass of M2∼1.2×108M⊙M_2 \sim1.2 \times10^{8}M_\odot and is shown by the thin line. The line color indicates the time in Gyr in the simulation; the MBH information is stored at a cadence of 1.7 Myr, while the galaxy information is stored at an average cadence of ∼110\sim110 Myr. The host galaxy has M⋆∼7.8×1010M⊙M_\star\sim7.8 \times10^{10}M_\odot and is captured at the closest timestep (t=7.67t = 7.67 Gyr) following the numerical merger of the MBHs at t=7.42t = 7.42 Gyr. The bottom panel zooms in to the central ∼3\sim3 kpc of the galaxy. In this instance, the numerical merger occured when the MBHs were separated by ∼0.47\sim0.47 kpc and their center-of-mass position was located ∼0.61\sim0.61 kpc from the galaxy center. The angle Θ\Theta between the orbital angular momentum and galaxy stellar angular momentum is Θ∼83∘\Theta\sim83^\circ, making this a corotating merger.

The aim of this work is not solely to provide another MBH merger rate estimate to the incredibly vast range of predictions, which span up to ∼6\sim6 orders of magnitude across simulations (Izquierdo-Villalba et al. 2026), but also to examine the prevalence of certain galaxy features, highlighted in HB25 to be conducive to MBH mergers, among a cosmological population of merger hosts. These are key questions surrounding the upcoming LISA mission, set to launch in 2035, which will be able to detect gravitational waves from MBHBs to extremely high redshift (z≳12z \gtrsim12, Amaro-Seoane et al. 2023).

The organization of the paper is as follows. We outline our technique in § 2, including an introduction to our post-processing model and base simulation, as well as details on how we determined galaxy properties from the simulation. § 3 discusses our results, compares them to other standard coalescence timescale models, and calculates the MBH merger rate for LISA. We discuss caveats of our work in § 4 and conclude in § 5.

Methods

In this section, we present a brief description of both the galaxy-informed timescale (2.1) and the Romulus25 cosmological simulation (2.2). We also describe how we sourced galaxy information from the galaxies containing MBH mergers within Romulus25, for both the galaxy-informed timescale and the timescales by SK15 and V15 (2.3).

Direct N-body simulations and galaxy-informed coalescence timescale

The galaxy-informed timescale used in this work is introduced in HB25. We briefly summarize the testing suite and timescale here.

HB25 collected 28 direct N-body simulations, accumulated over the course of a decade, of realistic galaxy centers hosting MBHBs. The galaxies were constructed in equilibrium and modeled after observations of real galaxies that vary in size, content, and structure. All galaxies exhibit some degree of triaxiality, and many contain NSCs and bulges. These galaxies are described in more detail in the original publications (Holley-Bockelmann & Khan 2015; Khan et al. 2015; Khan & Holley-Bockelmann 2021; Khan et al. 2024) and summarized in Table 1 of HB25.

The simulations focus on the stellar interactions with the MBHB by simulating only the central few hundred parsecs (depending on the model) of a gas and dark-matter (DM)-free galaxy. The secondary MBH is sent inwards on either planar prograde (Θ=0∘\Theta= 0^{\circ}) or retrograde (Θ=180∘\Theta= 180^{\circ}) orbits, where Θ\Theta is the angle between the orbital angular momentum and the stellar angular momentum of the host. The testing suite in HB25 varied cos⁡Θ\cos\Theta between -1 and 1.

HB25 performs a best-subset selection and linear regression over tens of parameters relating to the MBHB (e.g. total mass, initial eccentricity, mass ratio, etc.) and the galaxies (e.g., stellar mass, Sérsic index, NSC mass, etc.) to determine which few best describe the coalescence time (TcoalT_{\mathrm{coal}}) of the binary. The resulting four-term equation (Eq. (5) of HB25) with an R2=0.86R^{2} = 0.86 is

log⁡(Tcoal)=8.2+0.19log⁡(Mtot)−0.16log⁡(MNSC)+0.71(v/σproj)−0.32(Reff),(4)\log(T_{\mathrm{coal}}) = 8.2 + 0.19\log(M_{\mathrm{tot}}) - 0.16\log(M_{\mathrm{NSC}}) + 0.71(v/\sigma_{\mathrm{proj}}) - 0.32(R_{\mathrm{eff}}), \tag*{(4)}

where MtotM_{\mathrm{tot}} is the total mass of the MBHB, MNSCM_{\mathrm{NSC}} is the mass of the NSC (if the galaxy is nucleated), v/σv/\sigma is the ratio of the circular stellar velocity to the stellar velocity dispersion measured at the binary influence radius, v/σ∣proj=v/σcos⁡Θv/\sigma|_{\mathrm{proj}} = v/\sigma\cos\Theta denotes the projection of v/σv/\sigma onto the plane of the binary, and ReffR_{\mathrm{eff}} is the effective radius of the galaxy. HB25 finds that Eq. (4) reproduces the true coalescence times from the simulations to within ∼30%\sim30\%.

The Romulus25 simulation

We implemented the galaxy-informed timescale onto the MBHB population of Romulus25. Romulus25 (Tremmel et al. 2017) is a cosmological hydrodynamic simulation with a volume of (25cMpc)3(25\mathrm{cMpc})^{3} run to z=0z = 0 with the massively parallel Tree+SPH code, ChaNGa (Menon et al. 2015). ChaNGa builds upon physics modules used in the code Gasoline (Wadsley et al. 2004); those used to run Romulus25 include a cosmic UV background (Haardt & Madau 2012), star formation and “blastwave” supernovae feedback (Stinson et al. 2006) and low-temperature metal cooling (Guedes et al. 2011).

BHs in Romulus25 are seeded based upon physical properties of the gas in the simulation. Gas particles are converted into BH particles if they are quite metal poor (Z<3×10−4Z⊙Z < 3 \times10^{-4}Z_{\odot}), exceed 15 times the density required to undergo star formation (n>3cm−3n > 3\mathrm{cm}^{-3}), and fall within a temperature range which is cooler than the temperature required to form stars (9500K <T<< T < 10,000K), implying a collapse timescale that is quicker than the timescale for star formation. These conditions are consistent with seeding scenarios which require rapidly collapsing gas, including the channels for direct-collapse (Begelman et al. 2006; Lodato & Natarajan 2006), stellar runaway in dense clusters (Devecchi & Volonteri 2009; Davies et al. 2011) or even rapid accretion onto the remnant BH of a Population III star (Volonteri & Rees 2005; Inayoshi et al. 2016; Sassano et al. 2023). This prescription provides a more physical interpretation of the MBH occupation fraction compared to models that seed MBHs when a halo mass threshold is met (Sijacki et al. 2015; Schaye et al. 2015; Chen et al. 2022a). The seed mass for the resulting BHs in Romulus25 is MBH,seed=106M⊙M_{\mathrm{BH,seed}} = 10^{6}M_{\odot}. Figure 2 shows the MBH−M⋆M_{\mathrm{BH}} - M_{\star} relation for all galaxies in Romulus25 hosting central MBHs (purple) compared to those hosting MBH mergers (pink) across redshift. Many MBHs begin to be seeded and even merge at high redshift (z≳6z \gtrsim6), though we don’t display these because they occur in low mass galaxies (M⋆<107.5M⊙M_{\star} < 10^{7.5}M_{\odot}) that are excluded from our sample. The merging MBHs show a general agreement with the median MBH−M⋆M_{\mathrm{BH}} - M_{\star} of the nonmerging population, with a slight underrepresentation of high BH masses in the high M⋆M_{\star} regime. Generally, though, the merging population is not preferential to a certain subset of the MBH population in Romulus25.

Black hole mass to galaxy stellar mass relation for all galaxies with central MBHs compared to galaxies hosting merging MBHs

Figure 2. Black hole mass to galaxy stellar mass relation for all galaxies with central MBHs compared to galaxies hosting merging MBHs. We show the relation at high redshift (1.5≲z≲51.5 \lesssim z \lesssim5, middle panel), and low redshift (z≲1.5z \lesssim1.5, right panel). The colorbar indicates the logged counts of the nonmerging central MBHs in Romulus25, with the associated purple line indicating the median per stellar mass bin. The magenta scatter points show all primary MBH masses involved in MBH mergers and their associated stellar masses, with medians per bin indicated by the magenta line. We correct the galaxy stellar masses by a factor of 0.6 to account for the systematic observational underestimation of stellar mass compared to simulation (see text, Munshi et al. 2013). We add the observed MBH−M⋆M_{\rm BH}-M_\star relation by Schramm & Silverman (2013) to the low redshift panel as a dashed line (dotted when extrapolating beyond the sample masses).

Once seeded, the BHs are not pinned to the centers of their halos (e.g. (Sijacki et al. 2015; Schaye et al. 2015)) but rather their dynamics are modified with a sub-grid force correction term (Tremmel et al. 2015). This term is meant to approximate the dynamical friction that a MBH undergoes when traveling through a sea of stars of much lower mass, which cannot be properly resolved in most cosmological simulations due to the relatively similar masses of BH and star particles. Tremmel et al. (2015) finds that this dynamical friction term appropriately delays the sinking timescale of the MBH towards the halo center, decoupling the MBH merger rate from the galaxy merger rate (see also Chen et al. 2022a; Ma et al. 2023; Genina et al. 2024, for other examples of sub-grid dynamical friction prescriptions). This, combined with the physical seeding mechanism, leads to plenty of off-center, or “wandering” black holes, which are thought to be common in large galaxies (Bellovary et al. 2010; Ricarte et al. 2021).

Despite the high prevalence of these wanderers, we restrict our sample to MBHBs that are roughly central with their hosts, as the NN-body simulations of HB25 explicitly model galactic centers. We limit our sample to mergers where the remnant black hole in the closest following timestep has a galactocentric distance rgal<2kpcr_{\mathrm{gal}} < 2\mathrm{kpc}; we choose this distance as galaxy merger remnants can have particularly ill-defined centers and may be uncertain up to this distance. In the top panel of Figure 3, we show the distribution of galactocentric distances for the remnant black hole as measured in the timestep most closely following the numerical merging time. Just over 80% of the mergers in the sample occur within the central four softening lengths (4ϵsoft=1.4kpc4\epsilon_{\mathrm{soft}} = 1.4\mathrm{kpc}) of the galaxy center.

Cumulative fractions of remnant black hole distances from the galaxy center and MBHB orbital separations

Figure 3. Top: Cumulative fraction of distances between the remnant black hole and the galaxy center (rgalr_{\rm gal}) in the timestep most closely following the numerical merger time. We exclude mergers that have rgal>2 kpcr_{\rm gal} > 2\,\mathrm{kpc}. Bottom: Cumulative fraction of MBHB orbital separations (Δr\Delta r) at the time of numerical merger. Separations of two softening lengths (2ϵsoft=0.7 kpc2\epsilon_{\rm soft} = 0.7\,\mathrm{kpc}) and four softening lengths (4ϵsoft=1.4 kpc4\epsilon_{\rm soft} = 1.4\,\mathrm{kpc}) are marked by the dashed and dotted lines, respectively.

BHs in Romulus25 are allowed to numerically merge if the following criteria (Bellovary et al. 2011) are met:

  1. (i) The relative distance (Δr\Delta r) of the two BHs is within two softening lengths: Δr<2ϵsoft=0.7kpc\Delta r < 2\epsilon_{\mathrm{soft}} = 0.7\mathrm{kpc}

  1. (ii) The two BHs have low enough relative velocities (Δv\Delta v) and accelerations (Δa\Delta a) that they may be considered bound: 12Δv<Δa⋅Δr\frac{1}{2}\Delta v < \Delta a \cdot\Delta r

At Δr≤0.7kpc\Delta r \leq0.7\mathrm{kpc}, BHs in Romulus25 are still squarely in the dynamical friction phase, and we consider them a bound, close pair (Tremmel et al. 2018). The cumulative distribution of separations of the numerical mergers from Romulus25 is shown in the bottom panel of Figure 3. The peak at Δr=0.7kpc\Delta r = 0.7\mathrm{kpc} corresponds to numerical mergers that occurred as soon as the distance threshold was met.

Since the NN-body simulations of HB25 model the central few hundred pc of the galaxies, which is approximately the scale that Romulus25 cannot model, we choose not to model additional dynamical friction time to connect the two phases. We exclude mergers of BHs that merged within 10 Myr of their formation time or that were formed within 4ϵsoft4\epsilon_{\mathrm{soft}} of another BH. We start with a sample of 1867 mergers that fit these criteria. In the next section, we gather the relevant host galaxy properties, and our final sample of MBH mergers where all required galaxy properties were available amounts to 1457.

Sourcing galaxy properties

By design, the galaxy properties of Eq. (4) are relatively easy to gather from cosmological simulations. BH information in Romulus25 is saved with a finer time resolution than that of the larger simulation volume, thus properties like MtotM_{\rm tot} are saved at the moment of numerical merger regardless of the time of the simulation snapshots (or "timesteps"). For galaxy properties, we take the state of the galaxy at the timestep most closely following the merger time to approximate the post-merger remnant that the MBHB would inhabit, which ranges anywhere from almost instantaneous to ∼400\sim400 Myr after the numerical merger, with ∼80%\sim80\% occurring within 100 Myr. To accurately trace the location of the MBHB with respect to the galaxy center, we linearly interpolate along each dimension of the galaxy center’s motion to approximate its position between timesteps. Figure 1 shows the orbital motion of a selected MBHB around the galaxy center as projected on the xyxy and yzyz planes of the host. We also exclude mergers assigned to galaxies with log⁡(M⋆/M⊙)<7.5\log(M_\star/M_\odot)<7.5, as these are poorly resolved in Romulus25 and correspond to galaxies with too few (≲150\lesssim150) star particles. Throughout this section, we make use of the TANGOS (Pontzen & Tremmel 2018) and pynbody (Pontzen et al. 2013) libraries to source information from our MBH merger hosts.

Stellar kinematics: measured v/σv/\sigma

We calculate v/σv/\sigma for each merger using the stellar kinematic properties of the galaxy at the timestep following the numerical merger. The stellar velocity profile is calculated assuming circular orbits around the enclosed stellar and DM mass at each radius, binned every 100 pc. For velocity dispersion, we calculate the spherically-averaged, three-dimensional, enclosed velocity dispersion profile for each galaxy, also binned every 100 pc. For bins with too few stars (<5<5) to calculate velocity dispersion, we fall back to using gas and/or DM particles. We choose to measure v/σv/\sigma at 4ϵsoft=1.44\epsilon_{\rm soft}=1.4 kpc from the galaxy center, as σ\sigma is an enclosed quantity and dynamics only just become resolved at 2ϵsoft2\epsilon_{\rm soft} from the galaxy center. While we cannot resolve to scales of rinfr_{\rm inf} of the MBHB as requested by the galaxy-informed timescale, we expect that the approximation (v/σ)inf∼(v/σ)4ϵsoft(v/\sigma)_{\rm inf}\sim(v/\sigma)_{4\epsilon_{\rm soft}} only has a minor downstream effect on the coalescence timescale given the small range of measured v/σv/\sigma.

In Figure 4d, we show the measured distribution of v/σv/\sigma. All of our merger hosts show some degree of ordered rotation, with a minimum value of v/σ∼0.5v/\sigma\sim0.5 and a median of v/σ∼0.9v/\sigma\sim0.9. About 40% of our sample is rotation-dominated (v/σ>1.0v/\sigma>1.0), which differs from the expectation that merger hosts are mostly dispersion-dominated (Bournaud et al. 2004; Cox et al. 2006). The largest values (v/σ∼1.2v/\sigma\sim1.2) occur in the Milky-Way mass regime (M⋆∼1010−11M⊙M_\star\sim10^{10-11}M_\odot). Note that v/σv/\sigma here is not directly analogous to that observed in real galaxies, as we are not accounting for viewing angle or dust radiative transfer necessary for obtaining accurate line-of-sight velocities from cosmological simulations (e.g. Bardati et al. 2024b; Horlaville et al. 2026) (see § 4.2 for the caveats of this method).

Two-dimensional distributions of the ingredients for the galaxy-informed timescale versus stellar mass

Figure 4. Two-dimensional distributions of each of the ingredients for the galaxy-informed timescale (Eq. (4)) vs. the stellar mass of the host galaxy. The topmost panel shows the distribution of M⋆M_\star, and each of the rightmost panels shows the one-dimensional distribution of the ingredient for the timescale. a) Distribution of MtotM_{\rm tot} of the MBHB at the time of numerical merger by M⋆M_\star. b) Distribution of assumed NSC masses of galaxies assigned NSCs by M⋆M_\star. c) Distribution of effective radii of galaxies hosting MBH mergers by M⋆M_\star. d) Distribution of stellar velocity divided by velocity dispersion by M⋆M_\star, where v/σv/\sigma is measured at four times the gravitational softening length (4ϵsoft4\epsilon_{\rm soft}) from the center of the galaxy. e) Distribution of cos⁡Θ\cos\Theta by M⋆M_\star, where Θ\Theta is the angle between the galaxy stellar angular momentum and the MBHB orbital angular momentum.

Orbital configuration

We calculate the angle Θ\Theta between the MBHB orbital angular momentum (l⃗\vec{l}) and the galaxy’s stellar angular momentum (L⃗\vec{L}), where cos⁡Θ=L⃗⋅l⃗/∣L⃗∣∣l⃗∣\cos\Theta=\vec{L}\cdot\vec{l}/|\vec{L}||\vec{l}|. We measure L⃗\vec{L} by taking the mean of the inner 1.4 kpc (≲4ϵsoft\lesssim4\epsilon_{\rm soft}) of the spherically-averaged stellar angular momentum profile of the galaxy. We calculate l⃗=M1r⃗1×v⃗1+M2r⃗2×v⃗2\vec{l}=M_1\vec{r}_1\times\vec{v}_1+M_2\vec{r}_2\times\vec{v}_2 where r⃗1\vec{r}_1, r⃗2\vec{r}_2 and v⃗1\vec{v}_1, v⃗2\vec{v}_2 are the positions and velocities of the primary and secondary MBHs with respect to the MBHB center-of-mass at the moment of numerical merger. We define co-rotation (or prograde orbits) such that 0∘≤Θ≤90∘0^\circ\leq\Theta\leq90^\circ and counter-rotation (or retrograde orbits) such that 90∘<Θ≤180∘90^\circ< \Theta\le180^\circ. We show the measured distribution of cos⁡Θ\cos\Theta in Figure 4e, which resembles that of a uniform distribution of inclinations. This implies that ∼50%\sim50\% of the MBHBs in our sample are counterrotating with respect to their hosts. To determine if this was an effect of MBH dynamics becoming uncertain below the resolution limit, we also measured cos⁡Θ\cos\Theta by calculating l⃗\vec{l} at the moment the secondary MBH came within 2ϵsoft2\epsilon_{\rm soft} of the galaxy center. We found cos⁡Θ\cos\Theta to remain uniformly distributed, and conclude that this measurement is not particularly sensitive to the radius of measurement within the inner few kpc of the galaxy.

Nuclear star clusters

As NSCs are unresolved in Romulus25, we implement a very simple procedure to add NSCs onto (some) galaxies hosting MBH mergers. We assign a NSC to a merger host with a probability given by the occupation fraction constraints of Neumayer et al. (2020) (see their Fig. 3). While the occupation fraction differs slightly between early and late-type galaxies—high mass late-type galaxies host many more NSCs than early-types of the same mass—we leave a more detailed NSC assignment that distinguishes between galaxy types for a future work. We implement a scatter around the occupation fraction of 0.2 for M⋆≤108M⊙M_\star\le10^8M_\odot, 0.1 for 108M⊙≤M⋆<1010M⊙10^8M_\odot\le M_\star< 10^{10}M_\odot, and 0.4 for M⋆≥1010M⊙M_\star\ge10^{10}M_\odot, reflecting the large uncertainty in the presence of

NSCs in high mass galaxies. It may be true that MBH hosts are preferentially nucleated; Tremmel et al. (2024) finds the occupation fraction for MBHs in Romulus25 to match well with the observed occupation fraction of NSCs in the Local Volume (Hoyer et al. 2021; Carlsten et al. 2022). Still, we choose the more conservative approach to seed based on stellar mass instead of MBH presence.

For galaxies assigned a NSC, we allocate a mass given the scaling relation also from Neumayer et al. (2020):

log⁡MNSC=0.48log⁡(M⋆109M⊙)+6.51,(5)\log M_{\mathrm{NSC}} = 0.48 \log\left(\frac{M_{\star}}{10^{9}M_{\odot}}\right) + 6.51, \tag*{(5)}

with an intrinsic scatter of 0.6 dex. For galaxies not assigned a NSC, we omit the NSC term of Eq. (4) from the coalescence time, necessarily implying shorter coalescence times for nucleated galaxies (see § 4.1 for further discussion about this choice).

Of our MBH merger host sample, ∼50%\sim50\% are assigned a NSC. Their masses are displayed in Figure 4b. The masses span the range 104.3≲MNSC/M⊙≲108.910^{4.3} \lesssim M_{\mathrm{NSC}}/M_{\odot} \lesssim10^{8.9}, with a median NSC mass of MNSC∼106.7M⊙M_{\mathrm{NSC}} \sim10^{6.7}M_{\odot}. We are able to produce the full observed mass range of NSCs, which approximately spans from ∼104.5−109M⊙\sim10^{4.5} - 10^{9}M_{\odot} (Georgiev et al. 2016; Neumayer et al. 2020).

Effective radii

We calculate the effective radii of the Romulus25 MBH merger hosts by integrating over their simulated ii-band surface brightness profiles. We invoke a limiting surface brightness of μi≤32\mu_i \le32 and only fit over as far as the nearest satellite galaxy, if one exists in the halo. We make a cut for any very large effective radii (Reff>10kpcR_{\mathrm{eff}} > 10\mathrm{kpc}) as the galaxy-informed timescale (Eq. (4), HB25) begins to break down at these large scales. The resulting distribution of ReffR_{\mathrm{eff}} is shown in Figure 4c. Approximately ∼78%\sim78\% of our MBH merger hosts have Reff<3kpcR_{\mathrm{eff}} < 3\mathrm{kpc}, with a slight trend towards larger ReffR_{\mathrm{eff}} with larger M⋆M_{\star}. There is an overdensity in the M⋆∼1010M⊙M_{\star} \sim10^{10}M_{\odot} mass regime, with many cases of Reff∼1kpcR_{\mathrm{eff}} \sim1\mathrm{kpc}.

Extrapolated stellar properties at the influence radius

Although most of our terms involve bulk quantities, some information is required near the influence radius of the black hole, both for our case and for commonly-accepted prescriptions. To compare our timescale to SK15 and V15, we follow the procedure laid out in § 3.2 of SK15. First, we make the approximation that the binary spends most of its time at the transition separation, atra_{\mathrm{tr}}, between stellar hardening, which follows ∝a2\propto a^{2}, and GW hardening, which follows ∝a−3\propto a^{-3}. The combined hardening rate is then

dadt=dadt∣⋆+dadt∣GW=−Aa2−Ba3,(6)\frac{da}{dt} = \left.\frac{da}{dt}\right|_{\star} + \left.\frac{da}{dt}\right|_{\mathrm{GW}} = -Aa^{2} - \frac{B}{a^{3}}, \tag*{(6)}

where SK15 posits that A=SSK15A = S_{\mathrm{SK15}} as described by Eq. (1), and

B=64G3M1M2MtotF(e)5c2,(7)B = \frac{64G^{3}M_{1}M_{2}M_{\mathrm{tot}}F(e)}{5c^{2}}, \tag*{(7)}

where M1M_{1} and M2M_{2} are the mass of the primary and secondary, Mtot=M1+M2M_{\mathrm{tot}} = M_{1} + M_{2} is the total mass of the binary, and F(e)=(1−e2)−7/2(1+7324e2+3796e4)F(e) = (1-e^{2})^{-7/2}(1 + \frac{73}{24}e^{2} + \frac{37}{96}e^{4}) describes the evolution of orbital eccentricity (ee) during GW inspiral (Peters & Mathews 1963). We obtain the SK15 transition separation atr,SK15a_{\mathrm{tr,SK15}} by setting (da/dt)⋆=(da/dt)GW(da/dt)_{\star} = (da/dt)_{\mathrm{GW}}, which yields

atr,SK15=[64G2σinfM1M2MtotF(e)5c2]1/5.(8)a_{\mathrm{tr,SK15}} = \left[\frac{64G^{2}\sigma_{\mathrm{inf}}M_{1}M_{2}M_{\mathrm{tot}}F(e)}{5c^{2}}\right]^{1/5}. \tag*{(8)}

The timescale can then be approximated as:

tSK15=1SSK15atr,SK15=σinfGHρinfatr,SK15t_{\mathrm{SK15}} = \frac{1}{S_{\mathrm{SK15}}a_{\mathrm{tr,SK15}}} = \frac{\sigma_{\mathrm{inf}}}{G H \rho_{\mathrm{inf}} a_{\mathrm{tr,SK15}}}
≈15.8 Gyr(σinfkm s−1)(ρinfM⊙pc−3)−1(atr,SK1510−3pc)−1.(9)\approx15.8\ \mathrm{Gyr}\left(\frac{\sigma_{\mathrm{inf}}}{\mathrm{km\ s}^{-1}}\right)\left(\frac{\rho_{\mathrm{inf}}}{M_{\odot}\mathrm{pc}^{-3}}\right)^{-1}\left(\frac{a_{\mathrm{tr,SK15}}}{10^{-3}\mathrm{pc}}\right)^{-1}. \tag*{(9)}

For V15, the transition between stellar and GW hardening atr,V15a_{\mathrm{tr,V15}} occurs at the separation

atr,V15=(64G3M1M2Mtot5c2F(e)ahνσinfμSV15)1/(5+ν)(10)a_{\mathrm{tr,V15}} = \left(\frac{64G^{3}M_{1}M_{2}M_{\mathrm{tot}}}{5c^{2}F(e)}\frac{a_{\mathrm{h}}^{\nu}\sigma_{\mathrm{inf}}}{\mu S_{\mathrm{V15}}}\right)^{1/(5+\nu)} \tag*{(10)}

The timescale, then, is provided by

tV15=1SV15(atr,V15)atr,V15t_{\mathrm{V15}} = \frac{1}{S_{\mathrm{V15}}(a_{\mathrm{tr,V15}})a_{\mathrm{tr,V15}}}
≈1.7×108yr(rinf30pc)10+4ν5+ν(Mtot108M⊙)−5+3ν5+ν×μ−45+ν(4q(1+q)2)3ν−15+νF(e)−1+ν5+ν20ν.(11)\begin{aligned} \approx1.7 \times10^{8}\mathrm{yr}\left(\frac{r_{\mathrm{inf}}}{30\mathrm{pc}}\right)^{\frac{10+4\nu}{5+\nu}}\left(\frac{M_{\mathrm{tot}}}{10^{8}M_{\odot}}\right)^{-\frac{5+3\nu}{5+\nu}} \\ &\times\mu^{-\frac{4}{5+\nu}}\left(\frac{4q}{(1+q)^{2}}\right)^{\frac{3\nu-1}{5+\nu}}F(e)^{-\frac{1+\nu}{5+\nu}}20^{\nu}. \tag*{(11)} \end{aligned}

where SV15S_{\mathrm{V15}} is provided by Eq. (2). We take the fiducial values of μ=0.3\mu= 0.3 and ν=0.4\nu= 0.4 as provided by V15. For both timescales, we calculate F(e)F(e) by assuming a distribution of eccentricities to match the expectation for high eccentricities that typically result from cosmological simulations (Gualandris et al. 2022; Chen et al. 2022b; Francois et al. 2026). We assume a skewed Gaussian distribution centered roughly at e=0.8e = 0.8 with a skew factor a=−4a = -4 and scale of 0.25, and save a detailed study of eccentric MBHB orbits in Romulus25 for a future work.

To measure ρinf\rho_{\mathrm{inf}} for MBH merger hosts in Romulus25, we fit the slope of the stellar density profile using a power law index γ\gamma such that ρ∝r−γ\rho\propto r^{-\gamma}. We fit γ\gamma between 0.7 and 1.4 kpc from the galaxy center in attempt to describe the inner stellar structure while avoiding the region where the dynamics are not well resolved (<2ϵsoft< 2\epsilon_{\mathrm{soft}}). To calculate rinfr_{\mathrm{inf}}, we use the relation from Sesana (2010):

rinf=(3−γ)GMtotσinf2,(12)r_{\mathrm{inf}} = (3-\gamma)\frac{GM_{\mathrm{tot}}}{\sigma_{\mathrm{inf}}^{2}}, \tag*{(12)}

where we take σinf≈σ4ϵsoft\sigma_{\mathrm{inf}} \approx\sigma_{4\epsilon_{\mathrm{soft}}}. We choose 4ϵsoft=1.4kpc4\epsilon_{\mathrm{soft}} = 1.4\mathrm{kpc} as the radius of measurement as dynamical processes only just become resolved at 2ϵsoft=0.7kpc2\epsilon_{\mathrm{soft}} = 0.7\mathrm{kpc}, and σ\sigma measures an enclosed quantity up to the radius of measurement. Note that σinf\sigma_{\mathrm{inf}} here has the same value as the velocity dispersion used in our ν/σ\nu/\sigma measurements (Sec. 2.3.1).

We show the distribution of σinf\sigma_{\mathrm{inf}}, γ\gamma, ρinf\rho_{\mathrm{inf}}, and rinfr_{\mathrm{inf}} against the host galaxy stellar mass in Figure 5, excluding galaxies where the fit resulted in positive slopes (γ<0\gamma< 0) and capping our steepest slopes at γ≤3\gamma\le3. Figure 5a shows the relatively tight relation between stellar mass and stellar velocity dispersion, as expected, with increased dispersion in large galaxies. We find a median dispersion value in our sample of σinf∼130 km s−1\sigma_{\mathrm{inf}} \sim130\ \mathrm{km\ s}^{-1}. Our sample contains a median γ\gamma of ∼2.2\sim2.2, corresponding to a steep cusp. Figure 5b shows that a sizeable portion (∼28%\sim28\%) of the sample was capped at γ=3\gamma= 3; we choose to retain these cases rather than remove them as to not bias the entire sample toward smaller galaxy masses (M⋆≲109.5M⊙M_{\star} \lesssim10^{9.5}M_{\odot}). The rest of the sample shows an increase in cuspiness with higher M⋆M_{\star}, albeit with a large scatter. Figure 5c shows the influence radii for the merger hosts calculated using Eq. (12), ranging between 1 and 100 pc, with

Extrapolated galaxy properties for use in the timescales by SK15 and V15

Figure 5. Extrapolated galaxy properties for use in the timescales by SK15 and V15 vs. host galaxy stellar mass. The topmost panel shows the one-dimensional distribution of M⋆M_\star, and each of the rightmost panels shows the one-dimensional distribution of each of the following properties: a) Approximated stellar velocity dispersion at the MBHB influence radius (σinf\sigma_{\rm inf}) where we take σinf≈σ4ϵsoft\sigma_{\rm inf} \approx\sigma_{4\epsilon_{\rm soft}} vs. M⋆M_\star. b) Power-law index (γ\gamma) fit to the inner region (0.7–1.4 kpc) of the stellar density profile vs. M⋆M_\star. c) Influence radius (rinfr_{\rm inf}) of the binary defined as the radius enclosing twice the binary’s mass in stars, calculated using Eq. (12). d) Extrapolated stellar density at the MBHB influence radius (ρinf\rho_{\rm inf}) vs. M⋆M_\star. capped γ=3\gamma= 3 population contributing to the tail toward smaller rinfr_{\rm inf} values at high M⋆M_\star. Lastly, Figure 5d shows the extrapolated stellar densities at the influence radius, featuring two peaks in the distribution near ρinf∼100M⊙pc−3\rho_{\rm inf} \sim100M_\odot\mathrm{pc}^{-3} and ρinf∼105M⊙pc−3\rho_{\rm inf} \sim10^5M_\odot\mathrm{pc}^{-3}. Our largest ρinf\rho_{\rm inf} values reach to ∼109M⊙pc−3\sim10^9M_\odot\mathrm{pc}^{-3}, a similar maximum as found in other studies (e.g. Volonteri et al. 2020), though these are unphysically dense; NSCs, the densest stellar systems in the universe, can reach densities up to ∼105−6M⊙pc−3\sim10^{5-6}M_\odot\mathrm{pc}^{-3} at their centers (Pechetti et al. 2020; Hannah et al. 2024).

Since we take no steps to implement NSC cusps into the measured density profiles of our nucleated galaxies, we determine these density overestimates to be caused by poor stellar resolutions in the inner region of the galaxy, a limitation of all uniform-volume cosmological simulations. This implies the downstream calculations of TcoalT_{\rm coal} that rely on ρinf\rho_{\rm inf} and rinfr_{\rm inf} will be underestimates of the true coalescence time. This ρinf\rho_{\rm inf} measurement is also sensitive to the range over which the slope is fit; in a similar analysis of MBH merger hosts in the ASTRID simulation, Chen et al. (2022b) finds that measuring the slope out to r=1.5ϵsoftr = 1.5\epsilon_{\rm soft} versus r=2.0ϵsoftr = 2.0\epsilon_{\rm soft} manifests in an order-of-magnitude difference in ρinf\rho_{\rm inf} (though this effect would be less extreme in Romulus25, as the volume and resolution of ASTRID are much larger).

Results & Discussion

In this section, we present our distribution of coalescence times compared to those produced by the SK15 and V15 timescales, analyzed by host galaxy stellar mass and MBHB chirp mass (§ 3.1). In § 3.2, we analyze the influence of NSCs and counterrotation on the galaxy-informed timescale distribution. We also present the impact of the new coalescence times on the MBH merger rate and LISA event rate (§ 3.3), comparing with the merger rates produced by our comparison timescales, and situate these within the large range of existing merger rates predicted by cosmological simulations. Lastly, we discuss implications regarding host galaxy identification in the multi-messenger era under the assumption of short stellar scattering timescales (§ 3.4).

Coalescence time distributions and comparison to commonly used timescales

The probability distribution of coalescence timescales for a population of realistic galaxy hosts is shown in Figure 6. For the galaxy-informed time delay by HB25 that we implemented in this work, the distribution is centered at ∼108.2\sim10^{8.2} yr and spans ∼104.8\sim10^{4.8}–1010.210^{10.2} yr, the smallest range of the three timescales (∼5.4\sim5.4 dex). The shape is roughly Gaussian, with a slight indication of a double-peak that we identify in the next section as being related to the NSC assignment.

Distribution of coalescence timescales

Figure 6. Distribution of coalescence timescales for the galaxy-informed timescale by HB25 (purple) compared to that of SK15 (green) and V15 (blue). The median coalescence times for the three timescales, respectively, are 108.210^{8.2} yr, 108.610^{8.6} yr, and 108.910^{8.9} yr, respectively. When excluding hosts where the fit resulted in unphysically dense central stellar densities (ρinf>106M⊙\rho_{\rm inf} > 10^6M_\odot), the SK15 median becomes ∼108.9\sim10^{8.9} yr, and the V15 becomes ∼109.0\sim10^{9.0} yr.

To compare to other commonly implemented stellar hardening timescales, we also present the distributions of the SK15 and V15 timescales as applied to the Romulus25 merger hosts. The median of the SK15 distribution is ∼108.6\sim10^{8.6} yr, with the largest range of the three timescales (∼8.4\sim8.4 dex), including many cases that extend longer than the age of the Universe. The SK15 distribution features a distinct double-peak that directly follows that of the double-peak in ρinf\rho_{\mathrm{inf}} seen in Figure 5d; the coalescence times by SK15 that fall below ≲108.5\lesssim10^{8.5} yr correspond to MBH merger hosts with central densities ρinf≳103M⊙ pc−3\rho_{\mathrm{inf}} \gtrsim10^{3} M_{\odot}\,\mathrm{pc}^{-3}. Excluding unphysically dense systems (ρinf>106M⊙\rho_{\mathrm{inf}} > 10^{6} M_{\odot}) results in a SK15 median of ∼108.9\sim10^{8.9} yr. The V15 timescale yields a median of ∼108.9\sim10^{8.9} yr and ranges over ∼5.6\sim5.6 dex. This distribution does not feature the double-peaked shape, as it relies on ρinf\rho_{\mathrm{inf}} only implicitly. When excluding unphysically large densities, the median becomes ∼109\sim10^{9} yr. The ranges found here are comparable to those found by other cosmological simulations that have implemented the SK15 (e.g. Volonteri et al. 2020) and V15 (e.g. Chen et al. 2022b) timescales.

To further quantify the coalescence time distributions for the galaxy-informed timescale and the two comparison timescales, we show the median and 25/75th percentiles in the coalescence times vs. host galaxy stellar mass and MBHB chirp mass (Mc≡(M1M2)3/5/(M1+M2)1/5M_{\mathrm{c}} \equiv(M_{1}M_{2})^{3/5}/(M_{1}+M_{2})^{1/5}) in Figure 7. The galaxy-informed timescale is the quickest of the three across almost the whole M⋆M_{\star} and McM_{\mathrm{c}} ranges, only surpassed by SK15 in the range where extrapolated densities are highest (M⋆∼1010M⊙M_{\star} \sim10^{10}M_{\odot}, Mc∼107M⊙M_{\mathrm{c}} \sim10^{7}M_{\odot}). The galaxy-informed timescale is shortest in the highest mass bins (M⋆≳1011M⊙M_{\star} \gtrsim10^{11}M_{\odot}) as these correspond to systems with the highest ReffR_{\mathrm{eff}} and ν/σ\nu/\sigma, and MtotM_{\mathrm{tot}} values, even in the non-NSC cases. The V15 is the most constrained across M⋆M_{\star} and McM_{\mathrm{c}}, likely because the timescale relies on the power law index γ\gamma and subsequently-calculated rinfr_{\mathrm{inf}} (see Eq. (12)), which span a smaller range than the ρinf\rho_{\mathrm{inf}} values they produce. The SK15 is the most dependent on ρinf\rho_{\mathrm{inf}} and thus most greatly feels the effect of poor stellar resolution. This can be seen in low M⋆M_{\star} galaxies where the central stellar densities are lowest and likely only represented by tens to hundreds of star particles. Figure 7 shows how the SK15 predicts that mergers in these low M⋆M_{\star} galaxies would be exceedingly rare, with the median TcoalT_{\mathrm{coal}} taking much longer than the age of the Universe. Since we do expect MBH mergers to occur in low-mass galaxies (e.g. Volonteri et al. 2020; De Cun et al. [33]), especially at high redshift (where low-mass mergers are assembling the earliest massive galaxies (Bhowmick et al. 2025), it is possible that density-dependent models that work well in isolated galaxy environments with appropriate stellar resolution may unfairly penalize low stellar mass galaxies when applied to large volume cosmological simulations. This, in turn, may penalize mergers of low McM_{\mathrm{c}} (here, ≲107M⊙\lesssim10^{7} M_{\odot}), as these are hosted in low M⋆M_{\star} galaxies, which sit at the very center of the expected mass range detectable by LISA (Amaro-Seoane et al. 2023; Colpi et al. 2024).

Median and percentile spread in coalescence time versus remnant galaxy stellar mass and MBHB chirp mass

Figure 7. The median (solid line) and 25th and 75th percentile spread (shaded region) in the coalescence time for the galaxy-informed timescale (HB25, purple) and the timescales by SK15 (green) and V15 (blue) binned by remnant galaxy stellar mass (left panel) and MBHB chirp mass (right panel).

The median TcoalT_{\mathrm{coal}} produced here by the galaxy-informed timescale is approximately an order of magnitude faster than those produced by the comparison timescales, which fall closer to ∼1\sim1 Gyr when considering physically realistic systems. As it is the general understanding that stellar scattering depends nonlinearly on many different galaxy properties, some common approaches in studies of MBH evolution (of which detailed stellar hardening models fall outside of the scope) is to apply a ∼1\sim1 Gyr time delay across the board (e.g. Bellovary et al. 2025), or to predict that stellar scattering times are always negligible compared to dynamical friction times (e.g. Izquierdo-Villalba et al. 2026). Our work here supports the latter more than the former, though even the galaxy-informed timescale produces instances where scattering takes ∼1\sim1 Gyr or longer. We emphasize the relative ease of implementation of the timescale may be a selling point for these cases where a deep dynamical understanding of any one given system can be foregone for scattering times that globally reflect the influence of the galaxy hosts on the MBHB population.

Coalescence times vs. galactic and orbital properties

We are also interested in the individual effect of certain galaxy properties on TcoalT_{\rm coal} for the galaxy-informed timescale. In Figure 8, we show the galaxy-informed coalescence times binned by a) host galaxy stellar mass, b) whether or not the galaxy was assigned a NSC, and c) the nature of the MBHB rotation with respect to the stellar angular momentum of the host. Figure 8a divides the hosts into three stellar mass bins: low mass (M⋆∈10[7.5−9]M⊙M_\star\in10^{[7.5-9]}M_\odot), medium mass (M⋆∈10(9−10.5]M⊙M_\star\in10^{(9-10.5]}M_\odot), and high mass (M⋆>1010.5M⊙M_\star> 10^{10.5}M_\odot) galaxies. We see that the high M⋆M_\star hosts tend toward having shorter coalescence times, with a median Tcoal∼107.9T_{\rm coal} \sim10^{7.9} yr compared to the medium M⋆M_\star (∼108.3\sim10^{8.3} yr) and low M⋆M_\star (∼108.2\sim10^{8.2} yr) cases.

Distribution of galaxy-informed coalescence times

Figure 8. Distribution of the galaxy-informed coalescence times binned by a) stellar mass of the remnant galaxy, divided into low mass (M⋆∈10[7.5−9]M⊙M_\star\in10^{[7.5-9]}M_\odot), medium mass (M⋆∈10(9−10.5]M⊙M_\star\in10^{(9-10.5]}M_\odot), and high mass (M⋆>1010.5M⊙M_\star> 10^{10.5}M_\odot) galaxies, b) NSC assignment and c) corotation or counterrotation of the MBHB with respect to the host.

Figure 8b displays the effect of the NSC term on the coalescence times. The MBHBs in hosts assigned a NSC ∼13\sim13 times more quickly than those without; the NSC case has a median TcoalT_{\rm coal} of ∼107.6\sim10^{7.6} yr, and the no NSC case has a median of ∼108.8\sim10^{8.8} yr. We know from isolated merger models (Ogiya et al. 2020; Khan & Holley-Bockelmann 2021) and zoom-in simulations of galactic centers (Mukherjee et al. 2025) that NSCs are extremely efficient in accelerating MBHBs through the stellar scattering phase due to high stellar densities and tidal torques. Our result here serves to demonstrate that even a simple NSC implementation to account for this effect makes an appreciable difference to global stellar scattering times due to the high likelihood of NSCs in the galaxy stellar mass range that MBHs tend to occupy. For the hosts not assigned a NSC, the median Tcoal∼108.8T_{\rm coal} \sim10^{8.8} yr is similar to that produced by the comparison timescales; even when excluding unphysically dense systems (ρinf>106M⊙pc−3\rho_{\rm inf} > 10^{6}M_\odot\mathrm{pc}^{-3}), the medians come to within a factor of 2, indicating the NSC term to be one of the biggest differentiators of the galaxy-informed timescale. We remind the reader that these NSCs are not physically seeded and evolved, as NSCs are not resolved in Romulus25. We discuss the caveats of our NSC prescription in § 4.1.

Finally, we display the coalescence times divided into the corotating (0∘≤Θ≤90∘0^\circ\leq\Theta\leq90^\circ) and counterrotating (90∘<Θ≤180∘90^\circ< \Theta\leq180^\circ) cases in Figure 8c. The counterrotating cases have a median TcoalT_{\rm coal} of ∼107.8\sim10^{7.8} yr, ∼6\sim6 times faster than the corotating cases, with a median TcoalT_{\rm coal} of ∼108.6\sim10^{8.6} yr. The physical interpretation is that MBHBs that are counterrotating at the point of numerical merger in Romulus25 would go on to be pumped to high eccentricities by interaction with the retrograde stellar orbits, which extracts angular momentum efficiently (Sesana et al. 2011; Holley-Bockelmann & Khan 2015; Mirza et al. 2017). The counterrotating sample shown in Figure 8c also displays a double peak caused by the subsample that are assigned NSCs moving towards shorter TcoalT_{\rm coal}. Naturally, then, the fastest cases in the entire sample correspond to MBHBs whose hosts are assigned NSCs and which are counterrotating.

Merger rate and expected LISA detections

Since we attribute TcoalT_{\rm coal} as the time spent in the late dynamical friction phase and the stellar scattering phase, which occur at approximately the MBHB separations where Romulus25 binaries undergo numerical merger (see Figure 3), we added the respective coalescence times to the numerical merging times in the simulation. We can calculate the astrophysical MBH merger rate as in Chen et al. (2022b):

dNdzdt=d2n(z)dzdVcdVcdt1dz11+z(13)\frac{\mathrm{d}N}{\mathrm{d}z\mathrm{d}t} = \frac{\mathrm{d}^{2}n(z)}{\mathrm{d}z\mathrm{d}V_c}\frac{\mathrm{d}V_c}{\mathrm{d}t}\frac{1}{\mathrm{d}z}\frac{1}{1+z} \tag*{(13)}

where VcV_c is the comoving volume of the Romulus25 box at the time of merger, n(z)n(z) is number of MBH mergers occurring in a given redshift bin, and 1/(1+z)1/(1+z) transforms the time element into the observer frame. We take the finite-interval approximation that

d2n(z)dzdVc=N(z)ΔVcΔzsim(14)\frac{\mathrm{d}^{2}n(z)}{\mathrm{d}z\mathrm{d}V_c} = \frac{N(z)}{\Delta V_c \Delta z_{\rm sim}} \tag*{(14)}

where N(z)N(z) is the number of mergers occuring in a given redshift bin with width Δz\Delta z, and Vsim=(25cMpc)3V_{\mathrm{sim}} = (25\mathrm{cMpc})^3 is the full Romulus25 volume in comoving units.

In Figure 9, we show the merger rates for undelayed mergers and the mergers delayed by the the galaxy-informed timescale and the timescales by SK15 and V15. We determine the redshift at which the merger rate peaks for each timescale (zpeakz_{\mathrm{peak}}), finding zpeak∼4.1z_{\mathrm{peak}} \sim4.1 for the no delay case, zpeak∼3.3z_{\mathrm{peak}} \sim3.3 for the galaxy-informed timescale, and zpeak∼2.6z_{\mathrm{peak}} \sim2.6 for both the SK15 and V15 timescales. Generally, longer coalescence times should correspond to a zpeakz_{\mathrm{peak}} closer to zero; however, the vast range of TcoalT_{\mathrm{coal}} resulting from the SK15 timescale, with many taking longer than the age of the Universe, reduces the number of mergers drastically. This limits the effective distribution of TcoalT_{\mathrm{coal}} to those <1010< 10^{10} yr, under which the median coalescence time by SK15 becomes quicker (∼107.6\sim10^{7.6} yr) than that of V15 (∼108.9\sim10^{8.9} yr).

Merger rate of MBHs in Romulus25 with no time delay (orange), the galaxy-informed delay (purple), the delay by SK15 (green), and the delay by V15 (blue)

Figure 9. Merger rate of MBHs in Romulus25 with no time delay (orange), the galaxy-informed delay (purple), the delay by SK15 (green), and the delay by V15 (blue), binned into 20 redshift bins between z=0z = 0 and z=15z = 15.

To calculate the number of expected mergers (NexpN_{\mathrm{exp}}) detectable by LISA, we integrate over all redshifts and assume a 4-year LISA mission lifetime (Tobs=4yrT_{\mathrm{obs}} = 4\mathrm{yr}), which is the low-end estimate of LISA’s capabilities (Amaro-Seoane et al. 2023; Colpi et al. 2024). We find Nexp∼109N_{\mathrm{exp}} \sim109 for the undelayed mergers, Nexp∼46N_{\mathrm{exp}} \sim46 for the galaxy-informed timescale, Nexp∼22N_{\mathrm{exp}} \sim22 for the SK15 timescale, and Nexp∼36N_{\mathrm{exp}} \sim36 for the V15 timescale. The galaxy-informed timescale provides the most optimistic estimate for such LISA detections of the three timescales, though even this should still be considered an underestimate; MBH mergers with total masses as low as ∼105M⊙\sim10^{5}M_{\odot} will be observable by LISA (Colpi et al. 2024). These low-mass MBHs are expected to increase the merger rate, but fall below the resolution that is seeded in the Romulus25 simulation. Still, our event rate predictions fit squarely within the larger landscape of LISA event rate predictions, which span from ≲1\lesssim1 up to thousands (Izquierdo-Villalba et al. 2026).

For each merger, we calculate the sky-averaged signal-to-noise ratio (SNR) for a four-year mission using lisacode (Petiteau et al. 2008). We display the median fractional change in SNR from the no-delay SNR (ΔSNR/SNR\Delta\mathrm{SNR}/\mathrm{SNR}) in the zmerge−M⋆z_{\mathrm{merge}} - M_{\star} plane (left column) and zmerge−Mcz_{\mathrm{merge}} - M_{\mathrm{c}} plane (right column) for each timescale in Figure 10, where zmergez_{\mathrm{merge}} is the redshift at the time of numerical merger in the simulations. Longer coalescence times push mergers closer (or even beyond) z=0z = 0, resulting in a high fractional change in SNR from the no-delay case. This is visible in the middle panel, where the SK15 timescale displays high fractional SNR increases at low M⋆M_{\star} and low McM_{\mathrm{c}}, and cases delayed beyond z=0z = 0 are not shown. The bottom panel displays the V15 timescale, where fewer mergers are delayed beyond z=0z = 0, but long TcoalT_{\mathrm{coal}} at high redshift and in the low stellar mass range (see Figure 7) produce high fractional SNR changes. The fractional SNR increases for the galaxy-informed timescale, shown in the top panel, are the least dependent on M⋆M_{\star} and McM_{\mathrm{c}} of the three timescales, and generally small, with most mergers only ∼1.1−2\sim1.1 - 2 times louder than the no-delay SNR. All three timescales result in distributions with median SNRs near ∼103\sim10^3, extending as far as SNR ∼105\sim10^5; the small fractional SNR changes found by the galaxy-informed timescale then imply the use of the numerical merger population in a given simulation, provided that delays due to dynamical friction have been accounted for, may be sufficient to achieve SNR distributions for basic uses.

Median fractional change in signal-to-noise ratio across redshift, host stellar mass, and binary chirp mass for three timescales

Figure 10. Median fractional change in signal-to-noise ratio (SNR) across redshift of numerical merger, host stellar mass (M⋆M_\star, left column), and binary chirp mass (McM_c, right column) for the three timescales. Top: our galaxy-informed coalescence timescale (Eq. (4), HB25). Middle: timescale set by the Sesana & Khan (2015) formalism. Bottom: timescale prescribed by Vasiliev et al. (2015). The color bar indicates the log of the fractional change in SNR (assuming a 4-year LISA mission) compared to the no-delay case. The SK15 and V15 timescales induce high fractional SNR changes in the high redshift and low stellar mass hosts, with the typical merger being ∼10−100\sim10 - 100 times louder than the no-delay SNR in those regimes. The galaxy-informed timescale induces SNRs that are ∼1.1−2\sim1.1 - 2 times louder than the no-delay SNR, implying the use of the numerical merger population may be sufficient for basic SNR calculations (providing dynamical friction delays have been accounted for). The steeper color gradient in the middle panel is reflective of the much longer delay times experienced in lower-mass galaxies, and hints that some MBH mergers have been pushed to the future, as shown in Figure 9.

Implications for host galaxy identification

The relatively short stellar scattering times we find, if combined with conditions that foster fast MBH pairing and sinking, may be beneficial in the attempt to identify the host galaxy of merging MBHs. It is the current understanding that a substantial time delay between the galaxy merger and subsequent MBH merger implies the loss of any merger-induced morphological signatures due to relaxation (Volonteri et al. 2020; DeGraf et al. 2021; Bardati et al. 2024a). In a study using the IllustrisTNG simulations, DeGraf et al. (2021) finds that morphological disturbances in galaxies hosting high mass MBH mergers can remain apparent on the order of ∼500\sim500 Myr. Estimates of the dynamical friction timescale for MBH hosts vary greatly, from extremely efficient mergers (often due to NSCs) on the order ∼100\sim100 Myr (e.g. Ogiya et al. 2020; Mukherjee et al. 2025) to the main bottleneck at hundreds of Gyr (e.g. Volonteri et al. 2020; Ma et al. 2021; Bortolas et al. 2022).

In Romulus25, most MBHs are able to pair on the order of ∼\sim Gyr (Tremmel et al. 2018). This implies that even the very quick stellar scattering timescales found would not necessarily justify searching for tidal signatures from a galaxy merger. In a study of MBH merger host morphology within Romulus25, Bardati et al. (2024a) found that classical bulges may be more useful signposts for MBH merger hosts than tidal features like rings, tails, and streams (see however Izquierdo-Villalba et al. 2023). In any case, when using cosmological simulations to compare MBH merger hosts to the general galaxy population, it is important to not overestimate the delay time between galaxy and MBH merger, as bulk galaxy properties such as color, star formation rate, and morphology may be artificially erased.

Rather than looking for morphologies that tend to host MBH mergers in our simulations, we may wish to prioritize our searches to include galaxy features we know to assist MBH mergers. For example, in the local universe, we may wish to rank a nucleated galaxy higher in the ranking scheme. The electromagnetic signatures of counterrotating MBHBs are less clear, though orbital plane orientation is likely related to the orientations of the accretion disk and radio jets (Gualandris et al. 2012). Our findings that most merger hosts in Romulus25 exhibit some degree of ordered rotation suggests that kinematic information may be relevant to the search for LISA host galaxies (see also Bardati et al. 2024b; Horlville et al. 2026). Finally, stellar scattering may provide reason to search for stellar cores in galaxy centers; this is an active area of research (e.g. Dospoulou et al. 2021; Harris & Gültekin 2024) with intriguing observational prospects (e.g. McDonald et al. 2026). Our understanding of MBH host demographics is still evolving, as are our strategies for electromagnetic followup and archival searches for merger hosts of GW sources across the frequency spectrum (Petrov et al. 2024; Drake et al. 2025).

Caveats & Limitations

In § 3, we found that the galaxy-informed merger timescale allows for relatively quick stellar scattering times for the Romulus25 galaxies compared to other estimates. In this section, we discuss the limitations of our method.

Nuclear star cluster treatment

Our current model for the NSC assignment is crude in that we take the stellar mass at the timestep most closely following the MBH merger and compare that to the local universe occupation fraction by Neumayer et al. (2020), which almost surely does not hold at higher redshift where most of the our MBH mergers occur before applying delays (see the no delay case of Figure 9). The fact that we predict short delay times makes it that much more difficult to constrain the influence of NSCs at the peak in the MBH merger rate where NSC observations become sparse. Our method of obtaining masses for these NSC assignments is also flawed in that the M⋆−MNSCM_{\star} - M_{\rm NSC} scaling relations differ among galaxy type and almost certainly change with redshift as well (Neumayer et al. 2020; Hannah et al. 2024).

Our model also assigns a NSC for every merger instance, regardless of any previous MBH merger history in that galaxy. It would be preferable to evolve the mass of a NSC with its host throughout cosmic time semi-analytically (e.g. Antonini et al. 2015a,b; Kritos et al. 2025). It may be the case that NSCs are entirely disrupted by extremely efficient stellar scattering; Khan & Holley-Bockelmann (2021) find stellar mass deficits (MdefM_{\rm def}) up to 4 − 5 times the total mass of the MBHB, which is much larger than the more canonical picture which have Mdef≈0.5MtotM_{\rm def} \approx0.5M_{\rm tot} (Merritt (2006)). In future studies, we will track the evolution of NSCs under the influence of MBH mergers and investigate the impact of high MdefM_{\rm def} on global MBH merger timescales.

It is also important to acknowledge that the terms of Eq. (4) are by no means independent, and that omitting the NSC term without also modifying the other features is a slight statistical faux-pas. The presence of a NSC may also provide physical reason to modify certain unresolved processes in the simulation; for example, the nature of NSC rotation not only may signal something about its formation channel (Tsatsi et al. 2017; Pinna et al. 2021), but is also likely affected by the presence of the MBHB (Mastrobuono-Battisti et al. (2023)).

Kinematics and orbital setup

In Figure 4e, we presented a measured distribution of MBHB orbital plane inclinations with respect to the stellar angular momentum of each MBH merger host at the moment of numerical merger, and found the distribution to be roughly uniform. We found that this was also the case when measuring l⃗orb\vec{l}_{\rm orb} when the MBHB center-of-mass came within 2ϵsoft2\epsilon_{\rm soft} of the galaxy center, implying that the result is not sensitive to the radius of measurement within the central few kpc. It is not necessarily true that the inclinations would be isotropic on larger scales; satellite infalls towards larger galaxies are expected to be mainly radial, as they may be infalling along dark matter filaments (Benson 2005; Wetzel 2011; Li et al. 2020) and would then have preferential infall angles. For an appropriate comparison in Romulus25, we would need to ensure that each secondary MBH is associated with a subhalo; this isn’t necessarily true for every MBH merger in our sample (see the example galaxy in Figure 1) as BHs are seeded according to properties of local gas particles rather than placed at the centers of halos/subhalos.

The sense of rotation as described by cos⁡Θ\cos\Theta found itself in the galaxy-informed timescale because it helped to maximize the R2R^2 of the coalescence time fit, even surpassing the initial (eie_i) and final eccentricities (efe_f) as potential terms. In the simulations by HB25, cos⁡Θ\cos\Theta and efe_f were highly correlated, as counterrotating cases quickly had their eccentricities pumped to high values, a phenomenon observed in other numerical setups thought to be caused by many cumulative retrograde stellar encounters which bleed angular momentum from the binary (Iwasawa et al. 2011; Mirza et al. 2017). Because cos⁡Θ\cos\Theta described this phenomenon well for the galaxy-informed timescale, we chose not to perform a detailed study of eccentricities in the Romulus25 situation, saving this for a future investigation. We instead assume a high ee distribution for use in the SK15 and V15 timescales motivated by existing studies of MBHB eccentricities in cosmological simulations (Chen et al. 2022b; Francois et al. 2026).

In a recent work aiming to describe TcoalT_{\rm coal} using scaling relations of host galaxy properties, Francois et al. (2026) found that incorporating the eccentricity when the MBHB becomes bound (ebe_b) was necessary to achieve high predictive power for TcoalT_{\rm coal}. They argue that the choice to fold orbital information into the kinematic term of the timescale (v/σ∣proj=v/σcos⁡Θv/\sigma|_{\rm proj} = v/\sigma\cos\Theta) is what enables the high predictive power of Eq. (4), and that the coupling of v/σv/\sigma with cos⁡Θ\cos\Theta may not hold in systems with less rotational support. The v/σv/\sigma values that we display in Figure 4d are even higher (more rotationally supported) than those measured in HB25. While it’s possible that a more detailed v/σv/\sigma treatment may result in less rotational support (see v/σv/\sigma values found in Bardati et al. 2024b), the kinematic makeup of our hosts we find in this work implies that we remain in the regime where rotational support and orbital inclination are coupled.

Lastly, we are not tracking the evolution of cos⁡Θ\cos\Theta throughout through the dynamical friction phase within Romulus25, nor are we prescribing an evolution for cos⁡Θ\cos\Theta throughout the stellar scattering phase, despite the evidence that the orbital plane may flip from counterrotating to corotating after a few hardening times (Gualandris et al. 2012; Mirza et al. 2017). Instead, we report cos⁡Θ\cos\Theta for the MBH merger population only at the time of numerical merger (when the secondary MBHs stop existing in the simulation), which should approximately correspond to the initial infall angle of the secondary MBH in the simulations by HB25.

Subgrid model limitations

It is important to note that the subgrid model deployed here was based on a suite of direct NN-body simulations. Direct NN-body simulations are built to model galactic centers, and therefore certainly do not span the range of stellar structures that may influence the overall sinking timescale. Specifically, spiral arms and bars are lacking, and these have been shown to potentially stall the secondary MBH by capturing it in resonance orbits on the kiloparsec scale for ∼Gyr\sim\mathrm{Gyr}s (Bortolas et al. 2022; Herrero et al. 2025). The simulations were also constructed in equilibrium, which is not realistic for a post-merger galaxy remnant, especially if a NSC is deposited into a galaxy from a satellite merger. Even for the galactic center, the direct NN-body simulation suite did not model the most diffuse non-nucleated dwarf galaxies, which were shown to have exceptionally long merger timescale, if they were to host MBHs in the first place (e.g. Khan et al. 2024). However, perhaps the most critical missing component is gas, which has long been thought accelerate MBH mergers in most configurations (Mayer et al. 2007; Cuadra et al. 2009; Haiman et al. 2009; see however, D’Orazio & Duffell 2021; Siwek et al. 2023; Dittmann & Ryan 2024; Clyburn & Zrake 2026). The interplay between stellar and gas hardening is still an open question, though semi-analytic models (Bortolas et al. 2021), hydrodynamical simulations (Liao et al. 2024) and comparison studies (Kelley et al. 2017a; Izquierdo-Villalba et al. 2026) are making strides in this area.

Cosmological limitations

As with any cosmological simulation, there are inherent limitations from resolution, which manifests in missing low mass objects and physics on small scales, and from box size, which manifests in the limited ability to generalize results due to cosmic variance. We chose Romulus25 as our testbed because it is a uniform volume simulation that specializes in calculating accurate MBH orbits, necessary for the coalescence timescale prescription used here. However, it cannot resolve black holes with MBH<106M⊙M_{\rm BH} < 10^{6}M_{\odot} and galaxies with M⋆≲107.5M⊙M_{\star} \lesssim10^{7.5}M_{\odot}, the inclusion of which would increase our MBH merger rate and tip the typical MBHB mass ratio toward more unequal masses. The volume, Vsim=(25cMpc)3V_{\rm sim} = (25\mathrm{cMpc})^{3} also restricts the number of high mass BHs and galaxies, limiting the ability to extrapolate our results to the supermassive black hole range detectable by pulsar timing. This volume and mass range does, however, make Romulus25 an excellent test bed for MBHBs detectable by LISA, though it may overrepresent the true number of MBHs with exactly MBH=106M⊙M_{\rm BH}=10^{6}M_{\odot} due to the seeding prescription. In recent years, cosmological simulations have made headway in the implementation of low-mass seeds and multiple seeding channels ((Bhowmick et al. 2025; Koudmani et al. 2026; Prole et al. 2026)), as well as more physical seeding sites ((Dunn et al. 2018; Bhowmick et al. 2022; Prole et al. 2024)), which may raise the MBH merger rate as well (see (Izquierdo-Villalba et al. 2026, and references therein)).

Aside from the seeding prescription, other choices in the treatment of black hole/galaxy co-evolution may trickle down to bias the MBHB populations we see here. Feedback may be among the largest source of astrophysical uncertainty, manifesting in MBH mass, certainly, but also in galaxy stellar mass and central density ((Wright et al. 2026)), morphology, gas content, and color ((Barré et al. 2026)). In addition, these simulations, like most cosmological simulations, neglect black hole spin and gravitational wave recoil; black hole spin is known to change the feedback efficiency ((Nemmen et al. 2007; Tchekhovskoy et al. 2012; Ünal & Loeb 2020)), and recoil is expected to alter the MBH occupation fraction, growth history and bulge co-evolution ((Blecha et al. 2011; Izquierdo-Villalba 2026)). Finally, as we discussed previously, our seeding mechanism for NSC was done in post-processing, which bypasses a self-consistent treatment of MBH-bulge co-evolution ((Kormendy & Ho 2013)).

Conclusions

In this work, we applied a galaxy-informed timescale (Eq. (4), HB25) for the stellar scattering phase of MBH coalescence that depends on host galaxy properties to a population of MBH merger hosts in the Romulus25 simulation ((Tremmel et al. 2017)). We present our major takeaways below.

  • (i) The galaxy-informed merger timescale predicts a median TcoalT_{\rm coal} for the stellar scattering phase of ∼108.2\sim10^{8.2} yr, roughly an order of magnitude faster than timescales which rely on extrapolations of the stellar density profile. The range of TcoalT_{\rm coal} spans ∼104.8\sim10^{4.8}–1010.210^{10.2} yr for a population of 1457 merger hosts.

  • (ii) Approximately 50% of our MBH merger hosts were assigned a NSC; we predict these MBHBs would progress through stellar scattering ∼13\sim13 times faster than the non-NSC cases due to increased effective masses, tidal torques, and stellar densities that aid sinking and hardening (see (Ogiya et al. 2020; Khan & Holley-Bockelmann 2021; Mukherjee et al. 2025)).

  • (iii) At the time of numerical merger (where the BHs are separated by Δr≲0.7\Delta r \lesssim0.7 kpc, see Figure 3), MBHB orbital planes appear to be isotropically oriented with respect to their host galaxy disks. This means that 50% of our sample is counterrotating at the start of the stellar scattering phase; we predict these MBHBs would progress through stellar scattering ∼6\sim6 times faster than the corotating MB-HBs due to the early onset of strong GW emission aided by high eccentricities (see (Iwasawa et al. 2011; Holley-Bockelmann & Khan 2015; Mirza et al. 2017)).

  • (iv) All MBH merger host galaxies in our sample had some degree of ordered rotation near the center (4ϵsoft4\epsilon_{\rm soft}), with ∼40%\sim40\% being rotationally-dominated. This may be counterintuitive to a common expectation that dispersion increases in post-merger galaxies (e.g. (Bournaud et al. 2004; Cox et al. 2006)), and underlines the importance of taking galaxy rotation into account when modeling the MBHB stellar scattering phase. The kinematic distribution we see here is reflective of the prevalence of gas accretion over unequal mass mergers in the buildup of these MBHB hosts (e.g. [75 and references therein]).

  • (v) The galaxy-informed timescale predicts a MBH merger rate of ∼11.5\sim11.5 yr−1^{-1} with the peak MBH merger rate occurring at z∼3.3z \sim3.3. We predict ∼46\sim46 MBH merger events over a 4-year LISA mission, with the typical SNR in the thousands. These are optimistic predictions when compared to other timescales applied to the same MBHB population, though more work will be required to address the large variance seen across different simulations (see (Izquierdo-Villalba et al. 2026)).

The method we describe here is based on higher resolution simulation suites and is easy to implement, but it is certainly not the last word in modeling MBHB coalescence times. Future work for modeling galactic centers includes modeling realistic shapes, kinematics, and stellar populations, including stellar mass binaries and post-Newtonian gravity; astrophysically-motivated hydrodynamics, including AGN and stellar feedback; and a live potential to capture the interplay between gas, stars, and MBHs at the subparsec scale. As our understanding of MBHB evolution in high resolution studies continues to improve, so then should our post-processing and sub-grid modeling in cosmological environments. Practically, any models should strike the balance of capturing the relevant physics while maintaining relative ease of implementation. Future work in this area includes more robust semi-analytic and sub-grid models for NSCs to capture their global effect on MBH sinking and hardening timescales across cosmic time.

Acknowledgements

JSR would like to acknowledge M. Rizzo-Lopez, K. Hayes, Z. Hoelscher, R. Nowicki, and H. Blake-Goszyk for helpful comments on this manuscript. No AI was used in the making of this manuscript; AI was used sparingly in the debugging process. JSR acknowledges funding support from the National Science Foundation (NSF) EMIT Program (NSF2125764). This material is based upon work supported by Tamkeen under the NYUAD Research Institute grant CG014.

Data availability

A file containing the properties of the MBHBs and hosts collected here is available upon request, as well as the scripts used to generate figures for this paper. The Romulus25 database can be made available upon reasonable request. Romulus25 was run using ChaNGa which has a public version: https://github.com/N-BodyShop/changa. it can be placed in an Appendix which appears after the list of references.

This paper has been typeset from a TeX/LaTeX file prepared by the author.

References

  1. [33]De Cun V. I., Bellovary J. M., Katz M. L., 2023, MNRAS, 520, 3916
  2. [75]Lagos C. d. P., Schaye J., Bahé Y., van de Sande J., Kay S. T., Barnes D., Davis T. A., Dalla Vecchia C., 2018, MNRAS, 476, 4327
  3. [129]Yu Q., 2002, MNRAS, 331, 935

Paper details

vlnc:W_z7swtt6e
Version
v1 · latest
Published
Fresh — added to the Hub within the last 30 days
Added
License
CC BY 4.0
Usage

views · 11

downloads · 0

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

  • Physics and Astronomy

AI

View statement
Looking for related papers…
Contents