betaFeedback

WorksPhysics and Astronomy

The Broad-Line Region Size-Luminosity Relation in AGN Constrained by the Viewing Angle with the FRADO Model

Abstract

We investigate the Hββ broad-line region (BLR) size-luminosity (RBLRR_{\rm BLR}-L5100L_{5100}) relation of active galactic nuclei (AGN) using a Bayesian viewing angle inference framework based on the Failed Radiatively Accelerated Dusty Outflow (FRADO) model. First, we update the FRADO model by introducing a phase factor to account for the anisotropic visibility of the illuminated cloud surface. Second, we construct a Bayesian framework to constrain the viewing angle by matching the FRADO-based RBLRR_{\rm BLR} to the measured Hββ lags of 151 AGN from Wang & Woo (2024). We successfully reproduce the slope and intrinsic scatter of the observed RBLRR_{\rm BLR}-L5100L_{5100} relation, finding that the inferred viewing angles are broadly consistent with the estimates from dynamical modeling of velocity-resolved reverberation data, albeit with substantial scatter. As the observed Hββ time lag and line velocity depend on the viewing angle, the black hole mass (MBHM_{\rm BH}) calibrated from our work is typically larger than the reverberation-mapping estimates using an average virial factor. Our results provide a physically motivated interpretation of the RBLRR_{\rm BLR}-L5100L_{5100} relation and the systematic uncertainties in reverberation and single-epoch MBHM_{\rm BH} estimates.

Introduction

Supermassive black holes (SMBHs) at the centers of galaxies are known to play a significant role in galaxy evolution through feedback during the active galactic nucleus (AGN) phases [e.g. 19, 24, 22, 23]. Understanding AGN feedback and black hole physics requires reliable black hole mass (MBHM_{\mathrm{BH}}) measurements. Reverberation mapping (RM) provides MBHM_{\mathrm{BH}} for type-1 AGN based on the time lag between continuum and broad emission-line variability [10, 43], constraining a characteristic size of the broad-line region (RBLRR_{\mathrm{BLR}}).

One of the key results from the RM studies is the empirical relation between RBLRR_{\mathrm{BLR}} and continuum luminosity at 5100Å (L5100L_{5100}), which enables indirect MBHM_{\mathrm{BH}} estimates by inferring RBLRR_{\mathrm{BLR}} from L5100L_{5100} using a single-epoch spectrum [e.g. 27, 58, 46, 52, 7]. For example, M. C. Bentz et al. [8] reported a tight RBLRR_{\mathrm{BLR}}–L5100L_{5100} relation with a slope close to 0.5, which is consistent with simple photoionization models. This relation has been widely used for single-epoch MBHM_{\mathrm{BH}} estimates [e.g. 50, 45, 4, 48, 39, 47] (Q. Wu & Y. Shen 2022).

However, later studies showed that super-Eddington AGN tend to deviate from the RBLRR_{\mathrm{BLR}}–L5100L_{5100} relation [e.g. 30, 60] (P. Du et al. 2016; P. Du et al. 2018; S. S. Li et al. 2021), although the physical reason is not yet clear (see discussion by J. H. Woo et al. [60]). These results have motivated calibrations accounting for the Eddington ratio effect [see 57] (P. Du et al. 2018; P. Du & J. M. Wang 2019; L. M. Yu et al. 2020). By uniformly analyzing Hβ\beta lags for a large sample of AGN, S. Wang & J. H. Woo [54] recently updated the RBLRR_{\mathrm{BLR}}–L5100L_{5100} relation, reporting that the slope of the relation is ∼0.4\sim0.4 with an intrinsic scatter of ∼0.2\sim0.2 dex (see also J. H. Woo et al. [60]). J. H. Woo et al. [57] showed that three parameters, RBLRR_{\mathrm{BLR}}, L5100L_{5100}, and gas velocity, define a tight fundamental plane, suggesting the fundamental plane relation is more suitable for single-epoch mass estimates than the two-parameter RBLRR_{\mathrm{BLR}}–L5100L_{5100} relation.

Several theoretical models have been proposed to explain the origin of the BLR (e.g., [20, 18, 32, 12, 6, 5, 25 ) and the $R_{\mathrm{BLR}}$–$L_{5100}$ relation (see also {citep}][38, 32, 12, 5]. Among these models, the Failed Radiatively Accelerated Dusty Outflow (FRADO) model, first introduced by {citet}B. Czerny & K. Hryniewicz (2011), describes the BLR as a population of H$\beta$-emitting clouds lifted above the accretion disk by radiation pressure acting on dust, while gravity from the central black hole eventually causes the clouds to fall back after dust sublimation. The FRADO model predicts the inner and outer boundaries of the BLR, asymmetric emission-line profiles ({citet}M. H. Naddaf [33]), and the dependence of the virial factor on black hole mass ({citet}B. Czerny et al. [13]`).

The 2.5D FRADO model by M. H. Naddaf et al. [35] reproduces an RBLRR_{\mathrm{BLR}}–L5100L_{5100} relation similar to that reported by M. C. Bentz et al. [8] based on the particle-based three-dimensional implementation (see also [36, 26]. However, the 2.5D FRADO model does not explicitly include anisotropic visibility of the illuminated cloud surface to the line-of-sight (LOS), although such anisotropy has been suggested by BLR dynamical modeling studies ([42, 41, 56]). Since the far-side clouds may contribute more than near-side clouds to the transfer function used to define the representative time lag, a phase-dependent weighting scheme is needed to properly calculate the Hβ\beta time lag.

The viewing angle is another key factor in determining the LOS-dependent time lag. In practice, however, the viewing angle remains difficult to determine observationally for individual AGN, while several recent studies have attempted to constrain it using dynamical modeling of velocity-resolved RM data (e.g., [40, 42, 56, 55]) or interferometric phase spectra (e.g., [51, 1, 2, 3]).

With these motivations, we investigate whether the FRADO model, combined with a viewing angle inference framework, can reproduce the observed RBLRR_{\mathrm{BLR}}–L5100L_{5100} relation, by determining the representative time lag from the transfer function after correcting for the anisotropic visibility of each cloud. In our Bayesian framework, we constrain the viewing angle (θ\theta), which determines the projection effects on the observed time lag, line velocity, and MBHM_{\mathrm{BH}}. In § 2, we present the update of the 2.5D FRADO model. The Bayesian framework for the viewing angle calibration is presented in § 3. In § 4, we present the Bayesian inference results, including the inferred viewing angle and AGN properties. Discussion is presented in § 5, followed by summary and conclusions in § 6.

Time Lag constrained by the FRADO Model

In this section, we describe the update of the FRADO model and the calculation of Hβ\beta time lag. First, we describe the implementation of the BLR in the FRADO model (§ 2.1). Second, we present the phase factor to include anisotropic visibility of BLR clouds (§ 2.2). Third, we describe the grid model used to calculate time lag over a continuous parameter space (§ 2.3).

Calculation of Transfer Function

The BLR described by the FRADO model is an ensemble of non-interacting clouds, whose time-independent particle distribution is used to calculate the representative Hβ\beta time lag (τmodel\tau_{\mathrm{model}}). The particle distribution is determined by the black hole mass and luminosity (or equivalently, Eddington ratio λ\lambda), which set the gravitational potential of the central black hole and radiation pressure from the accretion disk, respectively. Once lifted from the disk surface, the clouds evolve under the combined effects of gravity and radiation pressure. The disk region contributing to the radiation pressure on each cloud is determined using the simplified shielding model introduced in M. H. Naddaf et al. [34] and M. H. Naddaf et al. [35]. When a cloud reaches the dust sublimation temperature, its dust is destroyed, radiation pressure ceases, and the cloud falls back toward the disk. Together with gravitational rotation, this vertical oscillatory motion produces the characteristic three-dimensional distribution of BLR clouds.

For a given BLR geometry specified by black hole mass and λ\lambda, the viewing angle (θ\theta) determines the LOS-dependent τmodel\tau_{\mathrm{model}}, which is calculated in three steps: computing the time lags of individual clouds, constructing the phase-weighted transfer function, and deriving τmodel\tau_{\mathrm{model}}, and extending the grid model to a continuous parameter space.

We start by calculating the time lag of individual clouds. The time lag of each cloud is defined as the light-travel path difference between the reprocessed lineemission and the continuum. Assuming that the continuum propagates directly from the central source to the observer, the time lag of each cloud is given by

τcloud=∥pcloud∥−pcloud⋅v^c,(1)\tau_{\mathrm{cloud}} = \frac{\lVert\mathbf{p}_{\mathrm{cloud}} \rVert- \mathbf{p}_{\mathrm{cloud}} \cdot\hat{\mathbf{v}}}{c}, \tag*{(1)}

where pcloud\mathbf{p}_{\mathrm{cloud}} is the cloud position vector relative to the central black hole, v^\hat{\mathbf{v}} is the LOS unit vector, and cc is the speed of light.

Then, we construct the transfer function by collecting the time lags of individual clouds.

Introduction of a Phase Factor to the Transfer Function

We introduce a phase factor into the transfer function to account for the anisotropic visibility of the illuminated side of each cloud. Following M. H. Naddaf et al. [36], we first assign each cloud an Hβ\beta emissivity determined by the incident photon flux. Then we multiply the emissivity by a dimensionless phase factor that accounts for the anisotropic visibility of the illuminated cloud surface. Because BLR clouds are expected to be optically thick to the central ionizing radiation, only the surface facing the central source is illuminated, and only part of this surface is visible to the observer (see Figure 1).

Illustration of the phase factor

Figure 1. Illustration of the phase factor (Equation (2)). The position vectors of three arbitrary clouds are denoted with red arrows. For each cloud, the illuminated surface due to the central source is shown in white, whereas the right panel shows the reflected surfaces to the LOS in white.

Assuming parallel incident rays along the p^cloud\hat{\mathbf{p}}_{\mathrm{cloud}} direction and spherical clouds, we define the phase factor as the fraction of the projected cloud surface visible to the observer that is illuminated by the central source as:

Phase≡1−p^cloud⋅v^2,(2)\mathit{Phase} \equiv\frac{1-\hat{\mathbf{p}}_{\mathrm{cloud}} \cdot\hat{\mathbf{v}}}{2}, \tag*{(2)}

where p^cloud\hat{\mathbf{p}}_{\mathrm{cloud}} is the normalized position vector of a cloud. Although FRADO clouds are point particles, we introduce a spherical shape to define the dimensionless phase

factor. The phase factor is unity when an observer sees the fully illuminated surface of a cloud, and zero when the illuminated surface is invisible along the LOS. As a result, clouds located on the far-side of the BLR from an observer have larger phase factors than those on the near-side, increasing the contribution of the far-side BLR to the transfer function (see also [42]) as illustrated in Figure 2. This effect of the phase factor becomes more pronounced with a larger θ\theta, as the projected separation of the two sides increases and the median of the transfer function shifts toward longer lags.

Example transfer functions of an AGN

Figure 2. Example transfer functions of an AGN with MBH=109.2M⊙M_{\mathrm{BH}}=10^{9.2}M_{\odot} and λ=0.16\lambda=0.16. The total transfer function (black line) is calculated using the Hβ\beta emissivity with and without the phase factor (dashed and solid lines, respectively). The corresponding far-side and near-side transfer functions are shown in red and blue, respectively.

We define τmodel\tau_{\mathrm{model}} as the centroid of the transfer function region above 70% of the peak. This definition yields time lags broadly consistent with the Gaussian-smoothed far-side peak adopted by M. H. Naddaf et al. [36]. Although τmodel\tau_{\mathrm{model}} differs from the cross-correlation function (CCF) lags commonly measured above 80% of the CCF peak in RM studies, it provides a similar characteristic BLR response time.

We present τmodel\tau_{\mathrm{model}} as a function of viewing angle for various pairs of MM and λ\lambda in Figure 3, showing that τmodel\tau_{\mathrm{model}} generally increases with viewing angle. In the following analysis, we will use the viewing angle dependent τmodel\tau_{\mathrm{model}} as a representative time lag of the BLR.

Viewing angle dependence of $\tau_{\mathrm{model}}$ for AGN with various $M$ and $\lambda$

Figure 3. Viewing angle dependence of τmodel\tau_{\mathrm{model}} for AGN with various MM and λ\lambda. τmodel\tau_{\mathrm{model}} is defined as the centroid above the 70% peak threshold (solid) or the far-side peak (dotted lines).

Extension to a Continuous Parameter Space

We extend the discrete FRADO lag grid to a continuous parameter space to evaluate τmodel\tau_{\mathrm{model}} for arbitrary input parameters of mass and Eddington ratio. The cloud distributions are computed on a two-dimensional grid spanning log⁡(M/M⊙)=6\log(M/M_\odot)=6–10 and log⁡λ=−2\log\lambda=-2–0, with τmodel\tau_{\mathrm{model}} evaluated over θ=0∘\theta=0^\circ–60∘60^\circ, appropriate for type 1 AGN. Rather than interpolating the full four-dimensional hypersurface (log⁡(M/M⊙),log⁡λ,θ,log⁡τmodel)(\log(M/M_\odot),\log\lambda,\theta,\log\tau_{\mathrm{model}}), we evaluate the grid at 0.5° intervals in θ\theta and perform interpolation or extrapolation in the (log⁡(M/M⊙),log⁡λ,log⁡τmodel)(\log(M/M_\odot),\log\lambda,\log\tau_{\mathrm{model}}) space at each fixed θ\theta. In the low-luminosity regime, some combinations of MM and λ\lambda fall below the incident ionizing photon-flux threshold adopted in the 2.5D FRADO model (M. H. Naddaf et al. [36]), 1018 cm−2s−110^{18}\ \mathrm{cm}^{-2}\mathrm{s}^{-1}. Consequently, the populated grid points form a triangular region in the (M,λ)(M,\lambda) plane rather than a complete rectangle.

For a set of input parameters (log⁡(M/M⊙),log⁡λ,θ)(\log(M/M_\odot),\log\lambda,\theta) with sufficient surrounding grid support, we estimate τmodel\tau_{\mathrm{model}} using local interpolation. We first identify the neighboring grid points along the log⁡(M/M⊙)\log(M/M_\odot) and log⁡λ\log\lambda directions and compute local linear estimates of log⁡τmodel\log\tau_{\mathrm{model}}. The two local estimates are then combined using an inverse-variance-weighted average to obtain the interpolated lag and its uncertainty.

If the input lies outside the populated region or lacks sufficient local grid support, we use plane-based extrapolation. In this case, we fit a plane to all populated grid points in the (log⁡(M/M⊙),log⁡λ,log⁡τmodel)(\log(M/M_\odot),\log\lambda,\log\tau_{\mathrm{model}}) space and evaluate it at the input values of log⁡(M/M⊙)\log(M/M_\odot) and log⁡λ\log\lambda. This treatment is justified because the FRADO time-lag surface at fixed θ\theta is nearly planar in the (log⁡(M/M⊙),log⁡λ,log⁡τmodel)(\log(M/M_\odot),\log\lambda,\log\tau_{\mathrm{model}}) space, with a typical rms scatter of ∼0.03\sim0.03 dex or smaller. This hybrid interpolation and extrapolation scheme provides τmodel\tau_{\mathrm{model}} throughout the continuous (M,λ)(M,\lambda) parameter space, including regions without sufficient local grid support.

Bayesian Framework for constraining the viewing angle

In this section, we present a Bayesian framework for inferring the viewing angle θ\theta by matching τmodel\tau_{\mathrm{model}} to the observed Hβ\beta lag (τobs\tau_{\mathrm{obs}}), for a given AGN with the observed RM mass (MobsM_{\mathrm{obs}}) and Eddington ratio (λobs\lambda_{\mathrm{obs}}). We apply this framework to a sample of RM AGN from S. Wang & J. H. Woo [54] in § 4.

For each trial value of θ\theta for an AGN with fixed MobsM_{\mathrm{obs}} and λobs\lambda_{\mathrm{obs}}, we obtain MM and λ\lambda as input parameters for the FRADO model and evaluate the likelihood of τobs\tau_{\mathrm{obs}} given τmodel\tau_{\mathrm{model}}. Our Bayesian framework is summarized in Figure 4. Note that MobsM_{\mathrm{obs}} and λobs\lambda_{\mathrm{obs}} represent the observed mass and Eddington ratio from the RM studies, while the unsubscripted quantities, MM and λ\lambda, denote the corresponding intrinsic parameters using the inferred θ\theta.

Bayesian framework used in this work

Figure 4. Bayesian framework used in this work. Given the observed quantities (MobsM_{\mathrm{obs}}, λobs\lambda_{\mathrm{obs}}, and τobs\tau_{\mathrm{obs}}), the framework uses the FRADO model and the viewing angle calibration of MobsM_{\mathrm{obs}} and λobs\lambda_{\mathrm{obs}} to infer θ\theta. Each trial parameter vector Φ\mathbf{\Phi} determines the relative probabilities of the trial viewing angles for each AGN. At each trial θ\theta, the corresponding MM and λ\lambda are calibrated and used to compute τmodel\tau_{\mathrm{model}}. The likelihood is evaluated by comparing τmodel\tau_{\mathrm{model}} with τobs\tau_{\mathrm{obs}} and marginalizing over θ\theta. Markov chain Monte Carlo (MCMC) is then used to sample Φ\mathbf{\Phi}. The posterior samples are propagated to infer the distributions of θ\theta, MM, λ\lambda, and τmodel\tau_{\mathrm{model}}.

θ\theta-Dependence of the Observed AGN Properties

The observed RM MBHM_{\mathrm{BH}} is defined as

Mobs=⟨f⟩cτobsΔVlos2/G,(3)M_{\mathrm{obs}}=\langle f\rangle c\tau_{\mathrm{obs}}\Delta V_{\mathrm{los}}^{2}/G, \tag*{(3)}

where τobs\tau_{\mathrm{obs}} is the Hβ\beta lag, ΔVlos\Delta V_{\mathrm{los}} is the Hβ\beta line width, and ⟨f⟩\langle f\rangle is an average virial factor. Since τobs\tau_{\mathrm{obs}} changes as a function of the viewing angle to the LOS, and ΔVlos\Delta V_{\mathrm{los}} corresponds to the LOS velocity, MobsM_{\mathrm{obs}} depends on θ\theta.

The virial factor of individual AGN also depends on the viewing angle since it reflects the BLR geometry and kinematics projected along the LOS. In fact, the viewing angle is likely to be a dominant source of variation in the virial factor [see also 41, 21, 56, 53]. However, an average virial factor is often adopted in practice due to the difficulty of constraining a virial factor for individual AGN [e.g. 59, 54]. This average virial factor may correspond to a representative viewing angle ⟨θ⟩\langle\theta\rangle of the adopted AGN sample, although the virial factor also depends on other BLR properties, such as the opening angle and radial extent of the BLR. The viewing angle dependence is also clear in the observed Eddington ratio defined as

λobs∝10×L5100,obsMobs.(4)\lambda_{\mathrm{obs}} \propto\frac{10 \times L_{5100,\mathrm{obs}}}{M_{\mathrm{obs}}}. \tag*{(4)}

The disk luminosity observed at 5100 Å (L5100,obsL_{5100,\mathrm{obs}}) is often calculated from the measured flux assuming isotropic emission, although the disk emission is anisotropic with respect to the LOS. By adopting a simplified thin Lambertian disk prescription, we find that the observed continuum luminosity scales with cos⁡θ\cos\theta due to the projected emitting area. However, the constant bolometric correction factor of 10 commonly applied to L5100,obsL_{5100,\mathrm{obs}} is calibrated for an AGN population and therefore implicitly corresponds to ⟨θ⟩\langle\theta\rangle. Thus, the bolometric luminosity and Eddington ratio of each AGN need to be calibrated using its individual viewing angle.

Constraining Representative Viewing Angle

We quantify the viewing angle effect on the observed parameters by parameterizing the difference between the inferred θ\theta and a representative viewing angle ⟨θ⟩\langle\theta\rangle under the assumption that only the viewing angle determines the virial factor. We determine the representative viewing angle ⟨θ⟩\langle\theta\rangle corresponding to the average virial factor ⟨f⟩\langle f\rangle in Equation (3). While ⟨f⟩\langle f\rangle is typically adopted from the calibration of the MBHM_{\mathrm{BH}}–stellar velocity dispersion relation by J. H. Woo et al. [59], individual estimates of ff and θ\theta are not available for the AGN used in the calibration. Instead, we use 37 velocity-resolved RM AGN from L. Villafaña et al. [53], for which ff is calculated as the ratio of the MBHM_{\mathrm{BH}} inferred from dynamical modeling to the observed virial product, and θ\theta is inferred from the same modeling. For the same sample, ⟨f⟩\langle f\rangle was constrained by S. Wang et al. [55]: ⟨log⁡f⟩=0.69±0.06\langle\log f\rangle= 0.69 \pm0.06 for the rms line dispersion and ⟨log⁡f⟩=0.58±0.06\langle\log f\rangle= 0.58 \pm0.06 for the mean line dispersion, which are almost equivalent to those constrained by J. H. Woo et al. [59]. Thus, adopting this value rather than the value from J. H. Woo et al. [59] does not significantly affect our results. For each AGN in the sample of L. Villafaña et al. [53], we generate paired Monte Carlo samples of ff and θ\theta. We then assign each Monte Carlo draw a Gaussian weight according to the deviation of its log⁡f\log f from the adopted ⟨log⁡f⟩\langle\log f\rangle, such that draws closer to ⟨log⁡f⟩\langle\log f\rangle receive larger weights. The weighted θ\theta draws from all AGN are then combined into a single distribution, from which we adopt the median ⟨θ⟩=28.8∘\langle\theta\rangle= 28.8^\circ as the representative viewing angle.

Modeling -Dependence

First, we model the viewing angle dependence of the time lag. Since the far-side BLR dominates the transfer function (§ 2.2), we assume the lag increases with the viewing angle as (1+sin⁡θ)(1+\sin\theta) for a thin BLR disk. By accounting for the shape of the transfer function (see Figure 2), we generalize this dependence as (1+βsin⁡θ)(1+\beta\sin\theta), where the dimensionless factor β\beta captures variations in the transfer function shape arising from the extended BLR geometry in the FRADO model. Then, the dependence of τmodel\tau_{\mathrm{model}} is expressed as

τmodel(θ)τmodel(⟨θ⟩)=1+βsin⁡θ1+βsin⁡⟨θ⟩.(5)\frac{\tau_{\mathrm{model}}(\theta)}{\tau_{\mathrm{model}}(\langle\theta\rangle)}=\frac{1+\beta\sin\theta}{1+\beta\sin\langle\theta\rangle}. \tag*{(5)}

Second, we adopt the prescription of S. Collin et al. [11] for the line velocity by combining rotational and turbulent components. If the rotation is dominant, the projected LOS velocity scales approximately as sin⁡θ\sin\theta and the corresponding mass scales as sin⁡2θ\sin^2\theta. In addition, we include a turbulent velocity component related to the BLR thickness (or opening angle). Following S. Collin et al. [11], we parameterize the ratio of the turbulent velocity to the local Keplerian rotational velocity with the BLR aspect ratio. We describe the aspect ratio as tan⁡ϕ\tan\phi, where ϕ\phi is the BLR half-opening angle, and introduce fturbf_{\mathrm{turb}} as a free parameter that determines the strength of the turbulent contribution to ΔVlos2\Delta V_{\mathrm{los}}^2. Therefore, ΔVlos2\Delta V_{\mathrm{los}}^2 scales as sin⁡2θ+fturbtan⁡2ϕ\sin^2\theta+f_{\mathrm{turb}}\tan^2\phi. Because the calibrated mass is inversely proportional to this factor, a nonzero fturbf_{\mathrm{turb}} prevents it from diverging when θ\theta approaches zero. We constrain fturbf_{\mathrm{turb}} through Bayesian inference and treat it as a global parameter common to all AGN in our Hβ\beta RM sample. We discuss the implications of this parameter further in Section 5.3.

Third, combining the dependences of τmodel\tau_{\mathrm{model}} and ΔVlos\Delta V_{\mathrm{los}}, we describe the θ\theta-dependence on mass as

MMobs=sin⁡2⟨θ⟩+fturbtan⁡2ϕsin⁡2θ+fturbtan⁡2ϕ⋅1+βsin⁡⟨θ⟩1+βsin⁡θ.(6)\frac{M}{M_{\mathrm{obs}}}=\frac{\sin^2\langle\theta\rangle+f_{\mathrm{turb}}\tan^2\phi}{\sin^2\theta+f_{\mathrm{turb}}\tan^2\phi}\cdot\frac{1+\beta\sin\langle\theta\rangle}{1+\beta\sin\theta}. \tag*{(6)}

Fourth, under a simplified thin Lambertian disk prescription, we assume that the observed continuum luminosity scales with cos⁡θ\cos\theta, which represents the projected emitting area. Therefore, the calibration of L5100L_{5100} is given as

L5100L5100,obs=cos⁡⟨θ⟩cos⁡θ.(7)\frac{L_{5100}}{L_{5100,\mathrm{obs}}}=\frac{\cos\langle\theta\rangle}{\cos\theta}. \tag*{(7)}

Finally, by combining Equations 6 and 7, we obtain the Eddington ratio prescription as

λλobs=cos⁡⟨θ⟩cos⁡θ⋅sin⁡2θ+fturbtan⁡2ϕsin⁡2⟨θ⟩+fturbtan⁡2ϕ⋅1+βsin⁡θ1+βsin⁡⟨θ⟩.(8)\frac{\lambda}{\lambda_{\mathrm{obs}}}=\frac{\cos\langle\theta\rangle}{\cos\theta}\cdot\frac{\sin^2\theta+f_{\mathrm{turb}}\tan^2\phi}{\sin^2\langle\theta\rangle+f_{\mathrm{turb}}\tan^2\phi}\cdot\frac{1+\beta\sin\theta}{1+\beta\sin\langle\theta\rangle}. \tag*{(8)}

We parameterize the opening angle ϕ\phi and shape factor β\beta as functions of the intrinsic AGN luminosity derived from the FRADO model as described in Section 3.4. Note that ϕ\phi is defined as the altitude range that includes 95% of the Hβ\beta emission.

Conditional Weights for Marginalization

We construct a conditional weight used to marginalize the likelihood over θ\theta using Equation 8, which relates λobs\lambda_{\mathrm{obs}} to λ\lambda through the θ\theta calibration. While θ\theta is not expected to correlate with MM or λ\lambda, the observed values, MobsM_{\mathrm{obs}} and λobs\lambda_{\mathrm{obs}} may depend on θ\theta as described in the previous section (see Equations (6) and (8)).

We build an intrinsic distribution of λ\lambda as an asymmetric split-normal distribution centered at log⁡λc\log\lambda_c, with left and right scatters σL\sigma_L and σR\sigma_R, respectively. For a fixed λ\lambda, (8) shows that the viewing angle calibration naturally produces an anti-correlation between λobs\lambda_{\mathrm{obs}} and θ\theta, because the ratio λ/λobs\lambda/\lambda_{\mathrm{obs}} increases monotonically with θ\theta for the positive values of fturbf_{\mathrm{turb}} and β\beta relevant to our model. Changing λ\lambda shifts this relation horizontally in the λobs\lambda_{\mathrm{obs}}–θ\theta plane. The parameters λc\lambda_c, σL\sigma_L, and σR\sigma_R therefore define the family of λobs\lambda_{\mathrm{obs}}–θ\theta relations allowed for the population where the central locus is set by λc\lambda_c.

The resulting parameter vector Φ\boldsymbol{\Phi} is

Φ=(λc,σL,σR,fturb).(9)\boldsymbol{\Phi}=(\lambda_c,\sigma_L,\sigma_R,f_{\mathrm{turb}}). \tag*{(9)}

We assign uniform priors over the ranges listed in Table 1.

ParameterPrior Range
λc\lambda_c(0.001,0.1)(0.001, 0.1)
σL\sigma_L(0,5)(0, 5)
σR\sigma_R(0,5)(0, 5)
fturbf_{\mathrm{turb}}(0,1)(0, 1)

Table 1. Uniform prior ranges adopted for the global parameters in the Bayesian inference.

For each sampled Φ\boldsymbol{\Phi}, we assign each AGN an object-specific conditional weight over the allowed θ\theta grid. The likelihood is then marginalized over this grid. For the iith AGN with an observed Eddington ratio λobs,i\lambda_{\mathrm{obs},i}, we define the intrinsic Eddington ratio λij\lambda_{ij} at the jjth viewing angle grid point using (8):

λij=λobs,igij⟨gij⟩cos⁡⟨θ⟩cos⁡θj1+βijsin⁡θj1+βijsin⁡⟨θ⟩.(10)\lambda_{ij}=\lambda_{\mathrm{obs},i}\frac{g_{ij}}{\langle g_{ij}\rangle}\frac{\cos\langle\theta\rangle}{\cos\theta_j}\frac{1+\beta_{ij}\sin\theta_j}{1+\beta_{ij}\sin\langle\theta\rangle}. \tag*{(10)}

where

gij≡sin⁡2θj+fturbtan⁡2ϕij,⟨gij⟩≡sin⁡2⟨θ⟩+fturbtan⁡2ϕij.(11)\begin{aligned} g_{ij} &\equiv\sin^{2}\theta_j+f_{\mathrm{turb}}\tan^{2}\phi_{ij},\\ \langle g_{ij}\rangle&\equiv\sin^{2}\langle\theta\rangle+f_{\mathrm{turb}}\tan^{2}\phi_{ij}. \tag*{(11)} \end{aligned}

Here, gijg_{ij} represents the combined rotational and turbulent contributions to ΔVlos2\Delta V_{\mathrm{los}}^{2} at θj\theta_j, while ⟨gij⟩\langle g_{ij}\rangle denotes the corresponding value at ⟨θ⟩\langle\theta\rangle. Moreover, using the FRADO cloud distributions, we calculate β\beta and ϕ\phi over the full (M,λ)(M,\lambda) grid and obtain linear fits as functions of the luminosity proxy lij=Mijλijl_{ij}=M_{ij}\lambda_{ij}:

ϕij (deg)=6.85log⁡lij−31.4,(12)\phi_{ij}\ (\mathrm{deg})=6.85\log l_{ij}-31.4, \tag*{(12)}
βij=0.196log⁡lij+0.115.(13)\beta_{ij}=0.196\log l_{ij}+0.115. \tag*{(13)}

For each λij\lambda_{ij} evaluated at θj\theta_j, we assign the iith AGN a (unnormalized) conditional weight wijw_{ij}. This is defined as

wij∝sin⁡θj{exp⁡[−12(log⁡λij−log⁡λcσL)2]∣dlog⁡λdθ∣θ=θj,if log⁡λij<log⁡λc,exp⁡[−12(log⁡λij−log⁡λcσR)2]∣dlog⁡λdθ∣θ=θj,if log⁡λij≥log⁡λc.(14)w_{ij}\propto\sin\theta_j \begin{cases} \exp\left[-\dfrac{1}{2}\left(\dfrac{\log\lambda_{ij}-\log\lambda_c}{\sigma_L}\right)^2\right]\left|\dfrac{\mathrm{d}\log\lambda}{\mathrm{d}\theta}\right|_{\theta=\theta_j}, & \text{if }\log\lambda_{ij}<\log\lambda_c,\\[1.2em] \exp\left[-\dfrac{1}{2}\left(\dfrac{\log\lambda_{ij}-\log\lambda_c}{\sigma_R}\right)^2\right]\left|\dfrac{\mathrm{d}\log\lambda}{\mathrm{d}\theta}\right|_{\theta=\theta_j}, & \text{if }\log\lambda_{ij}\ge\log\lambda_c. \end{cases} \tag*{(14)}

The factor sin⁡θj\sin\theta_j represents the isotropic viewing angle prior expected for randomly oriented systems in three-dimensional space [28]. Meanwhile, the Jacobian term ∣dlog⁡λ/dθ∣\left|\mathrm{d}\log\lambda/\mathrm{d}\theta\right| converts the population weight defined in log⁡λ\log\lambda space into the corresponding weight in θ\theta space. We normalize the weights as

w~ij=wij∑kwikΔθ,(15)\widetilde{w}_{ij}=\frac{w_{ij}}{\sum_k w_{ik}\Delta\theta}, \tag*{(15)}

so that

∑jw~ijΔθ=1.(16)\sum_j\widetilde{w}_{ij}\Delta\theta=1. \tag*{(16)}

Here, Δθ=0.5∘\Delta\theta=0.5^\circ denotes the grid spacing in θ\theta. For consistency with the FRADO model grid, we restrict the viewing angle to [0∘,60∘][0^\circ,60^\circ].

Marginalized Likelihood and Posterior

Using the conditional viewing angle weights defined in § 3.4, we construct the θ\theta-marginalized likelihood of τobs\tau_{\mathrm{obs}} and combine it with the priors on Φ\boldsymbol{\Phi} to obtain the posterior. For each AGN and viewing angle grid point θj\theta_j, we apply the calibration described in § 3.3 to obtain MM and λ\lambda from MobsM_{\mathrm{obs}} and λobs\lambda_{\mathrm{obs}}. These quantities, together with θj\theta_j, are then used to evaluate τmodel,i(θj)\tau_{\mathrm{model},i}(\theta_j) and its uncertainty. For each AGN ii and viewing angle grid point θj\theta_j, we define the likelihood by modeling log⁡τobs,i\log\tau_{\mathrm{obs},i} as a Gaussian variable centered on log⁡τmodel,i(θj)\log\tau_{\mathrm{model},i}(\theta_j),

log⁡τobs,i∣θj,Φ∼N(log⁡τmodel,i(θj),σij2).(17)\log\tau_{\mathrm{obs},i}\mid\theta_j,\boldsymbol{\Phi}\sim\mathcal{N}\left(\log\tau_{\mathrm{model},i}(\theta_j),\sigma_{ij}^{2}\right). \tag*{(17)}

where

σij2≡στ,obs,i2+στ,model,i2(θj).(18)\sigma_{ij}^{2}\equiv\sigma_{\tau,\mathrm{obs},i}^{2}+\sigma_{\tau,\mathrm{model},i}^{2}(\theta_j). \tag*{(18)}

Here, στ,obs,i\sigma_{\tau,\mathrm{obs},i} and στ,model,i(θj)\sigma_{\tau,\mathrm{model},i}(\theta_j) denote the uncertainties of τobs\tau_{\mathrm{obs}} and τmodel\tau_{\mathrm{model}} in log-space, respectively. The likelihood is then marginalized over the θ\theta grid,

p(log⁡τobs,i∣Φ)≈∑jw~ij p(log⁡τobs,i∣θj,Φ)Δθ(19)p(\log\tau_{\mathrm{obs},i}\mid\boldsymbol{\Phi})\approx\sum_j\widetilde{w}_{ij}\,p(\log\tau_{\mathrm{obs},i}\mid\theta_j,\boldsymbol{\Phi})\Delta\theta \tag*{(19)}

where w~ij\widetilde{w}_{ij} is the normalized conditional weight for the iith AGN at θj\theta_j. The conditional likelihood is given by

p(log⁡τobs,i∣θj,Φ)=12πσij2exp⁡(−rij22σij2),(20)p(\log\tau_{\mathrm{obs},i}\mid\theta_j,\boldsymbol{\Phi})=\frac{1}{\sqrt{2\pi\sigma_{ij}^{2}}}\exp\left(-\frac{r_{ij}^{2}}{2\sigma_{ij}^{2}}\right), \tag*{(20)}

where

rij≡log⁡τobs,i−log⁡τmodel,i(θj).(21)r_{ij}\equiv\log\tau_{\mathrm{obs},i}-\log\tau_{\mathrm{model},i}(\theta_j). \tag*{(21)}

The total log-likelihood of the sample is therefore

log⁡L(Φ)=∑ilog⁡p(log⁡τobs,i∣Φ).(22)\log\mathcal{L}(\boldsymbol{\Phi})=\sum_i\log p(\log\tau_{\mathrm{obs},i}\mid\boldsymbol{\Phi}). \tag*{(22)}

Combining the likelihood with the prior on Φ\boldsymbol{\Phi} yields the posterior distribution

p(Φ∣{log⁡τobs,i})∝p(Φ)L(Φ).(23)p(\boldsymbol{\Phi}\mid\{\log\tau_{\mathrm{obs},i}\})\propto p(\boldsymbol{\Phi})\mathcal{L}(\boldsymbol{\Phi}). \tag*{(23)}

In addition, we account for the measurement uncertainties in log⁡Mobs\log M_{\mathrm{obs}} and log⁡λobs\log\lambda_{\mathrm{obs}} by drawing 25 Gaussian realizations for each AGN and marginalizing the likelihood over these realizations. For each realization, we recompute the viewing angle calibration, the conditional viewing angle weights, and τmodel\tau_{\mathrm{model}}. The final likelihood is obtained by averaging the θ\theta-marginalized likelihood over the 25 realizations.

We sample the posterior using the affine-invariant ensemble sampler implemented in the emcee package. The resulting posterior samples constrain the four model parameters and are used to infer θ\theta of each AGN. We then propagate these posterior samples to obtain the corresponding posterior distributions of MM, λ\lambda, and posterior-predictive model time lag, τmodel\tau_{\mathrm{model}}.

Results

In this section, we present the main results obtained by applying the Bayesian framework to the Hβ\beta RM sample of S. Wang & J. H. Woo [54]. First, we describe the posterior constraints on the model parameters and the resulting distributions of θ\theta, MM, and λ\lambda for the Hβ\beta RM sample (§ 4.1). Second, we investigate the sample selection effect on the inferred θ\theta–λ\lambda and θ\theta–MM relations (§ 4.2). Third, we compare the inferred τmodel\tau_{\rm model} with the observed Hβ\beta lags and evaluate the reproduced RBLRR_{\rm BLR}–L5100L_{5100} relation against observations (§ 4.3).

Posterior Constraints

Applying the Bayesian framework to the Hβ\beta RM sample, we obtain the posterior constraints on the model parameters and the viewing angles as shown in Figure 5a. From the joint posterior distributions of the four parameters, we obtain the 1D parameter constraints as λc=0.013−0.0072+0.017\lambda_{\rm c}=0.013^{+0.017}_{-0.0072}, σL=0.89−0.48+1.59\sigma_{\rm L}=0.89^{+1.59}_{-0.48}, σR=0.67−0.17+0.17\sigma_{\rm R}=0.67^{+0.17}_{-0.17}, and fturb=0.046−0.012+0.012f_{\rm turb}=0.046^{+0.012}_{-0.012} with 1-σ\sigma uncertainties. Note that these posterior constraints depend on the input sample. Thus, these parameters are relevant for the Hβ\beta RM sample. By defining the maximum a posteriori (MAP) parameter vector, ΦMAP\boldsymbol{\Phi}_{\rm MAP}, that maximizes the posterior probability in the full four-dimensional parameter space, we obtain λc=0.0063\lambda_{\rm c}=0.0063, σL=0.36\sigma_{\rm L}=0.36, σR=0.78\sigma_{\rm R}=0.78, and fturb=0.046f_{\rm turb}=0.046. We adopt ΦMAP\boldsymbol{\Phi}_{\rm MAP} as the representative parameter vector for the fixed-Φ\boldsymbol{\Phi} analyses in Section 4.3, which assess the sensitivity of the results to uncertainty in Φ\boldsymbol{\Phi}.

Joint posterior distributions of the four model parameters and the observed Eddington ratio and inferred viewing angle

Figure 5. Left: Joint posterior distributions of the four model parameters. Regions enclosing 68%, 95%, and 99.7% are denoted as contours, while the marginalized 1D posterior distributions are shown in the diagonal panels. The 16th, 50th, and 84th percentiles of the 1D distributions are indicated by the blue dashed vertical lines. Right: 2D posterior probability distribution of the observed Eddington ratio (λobs\lambda_{\mathrm{obs}}) and the inferred viewing angle (θ\theta), showing the conditional distribution of θ\theta at a given λobs\lambda_{\mathrm{obs}}. Regions enclosing 16%, 50%, and 84% are denoted as contours.

Using the posterior constraints on Φ\boldsymbol{\Phi}, we infer the θ\theta distribution for each AGN in the Hβ\beta RM sample. We present the resulting family of λobs\lambda_{\rm obs}–θ\theta relations with a probability density map in Figure 5b. Despite the broad dispersion in θ\theta at fixed λobs\lambda_{\rm obs}, AGN with higher λobs\lambda_{\rm obs} preferentially have smaller viewing angles.

Then, we construct the joint posterior distributions of the inferred θ\theta and the calibrated MM and λ\lambda by drawing 200 samples of Φ\boldsymbol{\Phi} from the posterior and generating 200 θ\theta realizations for each sampled vector. For each θ\theta realization, we obtain the corresponding MM and λ\lambda through the viewing angle calibration (see § 3.3). We present the resulting distributions in Figure 6. Note that MM and λ\lambda are expected to be independent of θ\theta. However, the posterior contours exhibit residual trends with θ\theta, which we attribute to the incompleteness of the Hβ\beta RM sample (see § 4.2).

Joint posterior distributions of the viewing angle and intrinsic black hole mass and Eddington ratio

Figure 6. Joint posterior distributions of the viewing angle and the intrinsic MBHM_{\mathrm{BH}} (left) and Eddington ratio (right). Regions enclosing 16%, 50%, and 90% are denoted as contours. The centroid trends are indicated by the white lines.

For the Hβ\beta RM sample, we obtain a marginalized θ\theta distribution with a mean of 21.6∘21.6^\circ, which is smaller than the representative viewing angle ⟨θ⟩\langle\theta\rangle corresponding to ⟨f⟩\langle f\rangle (see Figure 7). It is also substantially smaller than the isotropic viewing angle prior adopted in our inference, p(θ)∝sin⁡θp(\theta)\propto\sin\theta. As the BLR is visible at relatively low viewing angles, this result is qualitatively consistent with the typical expected for type 1 AGN. We discuss this result further in Section 5.1.

Marginalized posterior probability distribution of $\theta$

Figure 7. Marginalized posterior probability distribution of θ\theta (black line), constructed using the sampling configuration of Figure 6. The 16th–84th percentile range is indicated by the shaded gray region, while the median is marked by the vertical dotted line. For comparison, the θ\theta distribution expected for an isotropic orientation in three-dimensional space, p(θ)∝sin⁡θp(\theta) \propto\sin\theta, is shown by the red line.

Selection Effects in the θ\theta–λ\lambda and θ\theta–MM Relations

Since θ\theta is expected to be independent of the intrinsic AGN properties, MM and λ\lambda, the weak correlations shown in Figure 6 require further investigation. We therefore test whether these trends arise from the limited dynamic range of the Hβ\beta RM sample. To test this interpretation, we generate a mock population of ∼5000\sim5000 AGN such that log⁡L5100,obs\log L_{5100,{\rm obs}} follows the same Gaussian distribution as the Hβ\beta RM sample, while log⁡λobs\log\lambda_{\rm obs} is uniformly distributed.

We compare the MobsM_{\rm obs} and λobs\lambda_{\rm obs} distributions of the mock AGN sample and the Hβ\beta RM sample in Figure 8a. The Hβ\beta RM sample lacks low-Eddington-ratio AGN at high luminosities and high-Eddington-ratio AGN at low luminosities, whereas the mock sample provides denser coverage of both regimes. With this more complete coverage in the λobs\lambda_{\rm obs} distribution, we now confirm no dependence of θ\theta on MM or λ\lambda, as shown in Figure 8b and Figure 8c. The residual correlations in Figure 6 are likely caused by the incomplete coverage of the Hβ\beta RM sample in the (λobs,Mobs)(\lambda_{\rm obs},M_{\rm obs}) plane. Note that the mock sample also shows relatively small θ\theta since we use the posterior samples of Φ\boldsymbol{\Phi} obtained from the Hβ\beta RM sample.

Joint distributions of observed quantities and inferred parameters

Figure 8. Left: Joint distribution of λobs\lambda_{\mathrm{obs}} and MobsM_{\mathrm{obs}} for the Hβ\beta RM sample, color-coded by the median inferred θ\theta. A mock population of 5000 AGN is shown as green points for comparison. Middle: Joint posterior distribution of MM and the inferred θ\theta. Right: Joint posterior distribution of λ\lambda and the inferred θ\theta. In the middle and right panels, the 16%, 50%, and 90% posterior-probability regions are shown as contours, and the centroid trends are indicated by the overplotted lines.

RBLRR_{\rm BLR}–L5100L_{5100} Relation

Using the sampling configuration described in § 4.1, we draw Monte Carlo samples of θ\theta for each AGN in the Hβ\beta RM sample. We apply the viewing angle calibration to each θ\theta sample to obtain the corresponding MM and λ\lambda, and then use (M,λ,θ)(M,\lambda,\theta) to calculate τmodel\tau_{\rm model} with the FRADO model (see § 2). The resulting samples define the posterior-predictive distribution of τmodel\tau_{\rm model} for each AGN.

As a consistency check, we compare τmodel\tau_{\rm model} with τobs\tau_{\rm obs} (Figure 9a), finding consistency between them with a negligible offset of 0.037 dex and an rms scatter of 0.22 dex, indicating that we properly reproduced the observed time lag. To test the effect of the uncertainty in Φ\boldsymbol{\Phi}, we fix the parameter vector at ΦMAP\boldsymbol{\Phi}_{\rm MAP} and generate 1000 viewing angle realizations. We find that the τmodel\tau_{\rm model} is consistent with τobs\tau_{\rm obs} with an offset of 0.042 dex and an rms scatter of 0.21 dex, indicating that uncertainty in Φ\boldsymbol{\Phi} has little effect on the result. We also repeat the analysis without the measurement uncertainties in MobsM_{\rm obs} and λobs\lambda_{\rm obs}, finding a comparable result.

Comparison of model and observed time lags and the reproduced observed radius-luminosity relation

Figure 9. Left: Comparison of τmodel\tau_{\mathrm{model}} with τobs\tau_{\mathrm{obs}}. The error bar of τmodel\tau_{\mathrm{model}} indicates the 68% credible intervals of the posterior-predictive τmodel\tau_{\mathrm{model}} distributions. The median inferred θ\theta of each AGN is color-coded. Right: The reproduced ‘observed’ RBLRR_{\mathrm{BLR}}–L5100L_{5100} relation at fixed ΦMAP\Phi_{\mathrm{MAP}} with the best-fit (red line) and the 1-σ\sigma uncertainty range obtained from 1000 θ\theta draws (red shaded region). The values of θ\theta and λobs\lambda_{\mathrm{obs}} are indicated by color and point size, respectively. The best-fit relation for the Hβ\beta RM AGN from S. Wang & J. H. Woo [54] (black line) and the M. C. Bentz et al. [8] relation (green dash-dotted line) are also presented for comparison.

Then, we test whether our model can reproduce the observed RBLRR_{\rm BLR}–L5100L_{5100} relation. We first fix the parameter vector at ΦMAP\boldsymbol{\Phi}_{\rm MAP} and generate 1000 Monte Carlo realizations of θ\theta for the Hβ\beta RM sample. For each realization, we measure the slope and intrinsic scatter and construct their probability distributions. We obtain an RBLRR_{\mathrm{BLR}}–L5100L_{5100} slope of 0.44−0.010+0.0110.44^{+0.011}_{-0.010} and an intrinsic scatter of 0.19−0.0085+0.00780.19^{+0.0078}_{-0.0085} dex using the 1000 realizations, where the uncertainties represent the 1-σ\sigma ranges, finding that the model properly reproduces the observed RBLRR_{\mathrm{BLR}}–L5100L_{5100} relation reported by S. Wang & J. H. Woo [54]. We repeat the analysis by drawing Φ\boldsymbol{\Phi} from its full posterior distribution using the same sampling configuration from § 4.1. We obtain an RBLRR_{\mathrm{BLR}}–L5100L_{5100} slope of 0.43−0.017+0.0160.43^{+0.016}_{-0.017} and an intrinsic scatter of 0.19−0.015+0.0160.19^{+0.016}_{-0.015} dex, which are in better agreement with those measured by S. Wang & J. H. Woo [54]. In Figure 9b, we illustrate the observed RBLRR_{\mathrm{BLR}}–L5100L_{5100} relation generated from a single θ\theta draw at fixed ΦMAP\boldsymbol{\Phi}_{\mathrm{MAP}}.

AGN with higher λ\lambda tend to lie below the mean RBLRR_{\mathrm{BLR}}–L5100L_{5100} relation, qualitatively consistent with the trend with observed Eddington ratio. However, the calibrated λ\lambda distribution of the Hβ\beta RM sample is systematically shifted toward lower values relative to the λobs\lambda_{\mathrm{obs}} distribution of S. Wang & J. H. Woo [54]. We discuss this further in § 5.4.

We successfully reproduce the shallower slope of the observed R-L relation and the residual trend toward smaller τobs\tau_{\mathrm{obs}} for AGN with higher λobs\lambda_{\mathrm{obs}} [54]. Note that this shallower slope is mainly caused by the viewing angle effect of the Hβ\beta RM sample, which has a specific Eddington ratio distribution. Although there is a self-correlation between τobs\tau_{\mathrm{obs}} and λobs\lambda_{\mathrm{obs}} since λobs∝L5100,obs/(τobsΔVlos2)\lambda_{\mathrm{obs}} \propto L_{5100,\mathrm{obs}}/(\tau_{\mathrm{obs}}\Delta V_{\mathrm{los}}^2), the offset from the intrinsic R-L relation is not expected to directly correlate with the intrinsic λ\lambda (see Figure 19 of J. H. Woo et al. [60]).

Discussion

In this section, we discuss the inferred viewing angle distribution of the Hβ\beta RM sample (§ 5.1), the dependence of our results on the adopted representative viewing angle ⟨θ⟩\langle\theta\rangle (§ 5.2), the treatment of the BLR opening angle ϕ\phi (§ 5.3), and the differences between the observed AGN properties and their calibrated counterparts (§ 5.4).

Viewing Angles of the Hβ\beta RM Sample

Our Bayesian inference applied to the Hβ\beta RM sample favors viewing angles smaller than the representative viewing angle ⟨θ⟩=28.8∘\langle\theta\rangle= 28.8^\circ, which was derived from the virial factors and viewing angles of individual AGN (see § 3.2). As shown in Figure 7, the viewing angle distribution of the sample is skewed toward low viewing angles compared to the isotropic expectation. This trend is also evident in comparison with θ\theta constrained by dynamical modeling. Overall, the inferred viewing angles of the Hβ\beta RM sample are lower than those typically expected for type 1 AGN.

Several studies have suggested that narrow-line Seyfert 1 galaxies (NLS1s), which typically show high Eddington ratios, may preferentially be viewed at low viewing angles [31, 37, 9, 14, 44] (T. Sarrabata et al. 2018). Although the majority of AGN in the Hβ\beta RM sample have Hβ\beta FWHM values greater than 2000 km s−1^{-1} and therefore are not classified as NLS1s, the proposed face-on interpretation of high-Eddington-ratio NLS1s is qualitatively analogous to the anticorrelation between λobs\lambda_{\mathrm{obs}} and θ\theta (Figure 5b).

Compared to the viewing angles from dynamical modeling reported by L. Villafaña et al. [53], the inferred θ\theta is lower by 8.5∘±13∘8.5^\circ\pm13^\circ for the 31 overlapping AGN. The two estimates are broadly consistent within their uncertainties. However, several objects have inferred viewing angles substantially smaller than the corresponding dynamical modeling estimates.

These differences arise because the inference adjusts θ\theta to bring τmodel\tau_{\mathrm{model}} into agreement with τobs\tau_{\mathrm{obs}}. When θ<⟨θ⟩\theta< \langle\theta\rangle, the viewing angle calibration increases MM and decreases λ\lambda relative to MobsM_{\mathrm{obs}} and λobs\lambda_{\mathrm{obs}}, respectively (see § 3.3). Because τmodel\tau_{\mathrm{model}} depends strongly on MM (see Figure 3), the increase in MM produces a longer τmodel\tau_{\mathrm{model}}. Thus, objects whose initial τmodel\tau_{\mathrm{model}} based on (Mobs,λobs,⟨θ⟩)(M_{\mathrm{obs}}, \lambda_{\mathrm{obs}}, \langle\theta\rangle) is much shorter than τobs\tau_{\mathrm{obs}} favor θ\theta smaller than ⟨θ⟩\langle\theta\rangle. Because this behavior depends on the adopted viewing angle calibration, the current calibration model may require further refinement. In particular, the viewing angle dependence of the line velocity should be modeled more realistically using the FRADO cloud distribution. We defer further calibration to future work.

The discrepancy between the inferred θ\theta and the dynamical modeling estimates may also arise from the degeneracy between θ\theta and other BLR parameters, particularly the opening angle ϕ\phi. In our Bayesian framework, a large ϕ\phi weakens the viewing angle dependence of the model time lag as well as the line velocity (see § 3.3), thereby broadening the probability distribution of θ\theta. The preference for small viewing angles is presumably due to the relatively small ϕ\phi predicted by the FRADO model: for the Hβ\beta RM sample, ϕ\phi rarely exceeds 30°, even for the most luminous AGN (see § 5.3). The degeneracy among the inferred θ\theta, MM, and ϕ\phi has been reported in BLR dynamical modeling (see also A. Pancoast et al. [41]; C. J. Grier et al. [21]). In dynamical modeling, this degeneracy produces a positive correlation between θ\theta and ϕ\phi in the joint posterior, allowing both parameters to shift toward either smaller or larger values. Consistent with this interpretation, AGN for which the inferred viewing angles are substantially smaller than the dynamical modeling estimates gener-

ally have large opening angles inferred from dynamical modeling.

In dynamical modeling, ϕ\phi is treated as a free parameter. In contrast, we calculate ϕ\phi from the FRADO cloud distribution specified by MM and λ\lambda (see § 3.4). This prescription yields a narrower range of ϕ\phi with typically smaller values than those inferred from dynamical modeling. Thus, it is necessary to further study the degeneracy between θ\theta and ϕ\phi to confirm the relatively low viewing angles inferred for the Hβ\beta RM sample. We further discuss the opening angle implementation in § 5.3. Although observationally challenging, spectropolarimetric constraints on the viewing angle, once available for individual AGN in a larger RM sample, will provide an important independent test of the inferred viewing angle distribution.

Representative Viewing Angle

The inferred θ\theta of individual AGN slightly depends on the choice of ⟨θ⟩\langle\theta\rangle, for which we adopt 28.8∘28.8^\circ based on the 37 AGN with dynamical modeling results (see § 3.2). As a consistency check, we derive an alternative ⟨θ⟩\langle\theta\rangle assuming an isotropic θ\theta distribution (see also [28]). For randomly oriented AGN, the viewing angle probability density, p(θ)p(\theta), follows p(θ)∝sin⁡θp(\theta) \propto\sin\theta, where p(θ)p(\theta) is the probability density per unit viewing angle. Assuming that type 1 AGN are observed within θ≤60∘\theta\leq60^\circ, the mean of this distribution is ∼39.2∘\sim39.2^\circ. Using this distribution, we define the average log virial factor as

⟨log⁡f⟩=∫060∘log⁡f(θ)sin⁡θ dθ∫060∘sin⁡θ dθ.(24)\langle\log f\rangle= \frac{\int_{0}^{60^\circ} \log f(\theta)\sin\theta\,d\theta}{\int_{0}^{60^\circ} \sin\theta\,d\theta}. \tag*{(24)}

and determine ⟨θ⟩\langle\theta\rangle from

log⁡f(⟨θ⟩)=⟨log⁡f⟩,(25)\log f\left(\langle\theta\rangle\right) = \langle\log f\rangle, \tag*{(25)}

where the viewing angle dependence of the virial factor is expressed as

f(θ)∝[sin⁡2θ+fturbtan⁡2ϕ](1+βsin⁡θ)−1.(26)f(\theta) \propto\left[\sin^{2}\theta+ f_{\mathrm{turb}}\tan^{2}\phi\right]\left(1+\beta\sin\theta\right)^{-1}. \tag*{(26)}

This expression follows from the mass calibration in Equation (6). From Equations (12) and (13), we obtain approximate ranges of ϕ∼10∘\phi\sim10^\circ–25∘25^\circ and β∼1.3\beta\sim1.3–1.71.7 over the typical range of log⁡(λM)∼6\log(\lambda M) \sim6–88 in the Hβ\beta RM sample. Furthermore, our Bayesian inference constrains fturbf_{\mathrm{turb}} to ∼0.04\sim0.04–0.060.06 across ⟨θ⟩∼30∘\langle\theta\rangle\sim30^\circ–40∘40^\circ. For the adopted ranges of ϕ\phi, β\beta, and fturbf_{\mathrm{turb}}, the corresponding representative viewing angle varies only from ⟨θ⟩∼35.0∘\langle\theta\rangle\sim35.0^\circ to 35.5∘35.5^\circ, indicating weak sensitivity to these inputs. We therefore adopt ⟨θ⟩=35∘\langle\theta\rangle= 35^\circ for the alternative configuration.

Using this alternative ⟨θ⟩=35∘\langle\theta\rangle= 35^\circ, we repeat the Bayesian inference and compare the posterior-predictive τmodel\tau_{\mathrm{model}} with τobs\tau_{\mathrm{obs}}. We find consistency between τmodel\tau_{\mathrm{model}} and τobs\tau_{\mathrm{obs}}, with an offset of 0.041±0.220.041 \pm0.22 dex and an intrinsic scatter of 0.100.10 dex.

Next, we compare the inferred θ\theta for ⟨θ⟩=35∘\langle\theta\rangle= 35^\circ with the dynamical modeling estimates, as in Section 5.1. We obtain an offset of −2.7∘±11∘-2.7^\circ\pm11^\circ, showing better agreement than for ⟨θ⟩=28.8∘\langle\theta\rangle= 28.8^\circ. As future Hβ\beta RM samples become larger and less affected by selection effects, the representative viewing angle adopted in our Bayesian framework may approach ⟨θ⟩=35∘\langle\theta\rangle= 35^\circ.

Lastly, we revisit the inferred viewing angle distribution of the Hβ\beta RM sample for the alternative choice of ⟨θ⟩=35∘\langle\theta\rangle= 35^\circ. The marginalized θ\theta distribution shifts toward larger viewing angles. Nonetheless, the resulting θ\theta distribution still shows a substantially larger fraction at small viewing angles than the isotropic distribution.

Opening Angle

In our Bayesian framework, ϕ\phi modulates the viewing angle dependence of ΔVlos2\Delta V_{\mathrm{los}}^{2} through the turbulent contribution fturbtan⁡2ϕf_{\mathrm{turb}}\tan^{2}\phi, where the free parameter fturbf_{\mathrm{turb}} controls its amplitude (see § 3.3). The inferred fturbf_{\mathrm{turb}} is ∼0.045\sim0.045 (∼0.056\sim0.056 for ⟨θ⟩=35∘\langle\theta\rangle= 35^\circ), implying that the opening angle ϕ\phi contributes to the turbulent-to-rotational velocity ratio with a coefficient of fturb∼0.21\sqrt{f_{\mathrm{turb}}} \sim0.21–0.240.24. Over the typical ranges of θ\theta and ϕ\phi for the Hβ\beta RM sample, the rotational component remains dominant, while the turbulent component associated with ϕ\phi is subdominant but non-negligible, weakening the viewing angle dependence of the line velocity. This result is qualitatively consistent with C. J. Grier et al. [21], who reported that the viewing angle has a greater effect on the line profile shape than the opening angle.

As discussed in Section 5.1, the degeneracy between θ\theta and ϕ\phi is an important source of uncertainty in the inferred θ\theta. In this work, we calculate ϕ\phi from the angular extent of the FRADO cloud distribution (see § 3.3). The resulting opening angles rarely exceed 30∘30^\circ, even for the most luminous AGN in the Hβ\beta RM sample. In contrast, the opening angles inferred from dynamical modeling span approximately 10∘10^\circ–70∘70^\circ and are typically larger than those predicted by the FRADO model. In both our inference and BLR dynamical modeling, the degeneracy between θ\theta and ϕ\phi produces a positive correlation, such that a larger ϕ\phi is associated with a larger θ\theta (see Section 5.1). The limited range of ϕ\phi in our model may therefore contribute both to the overall preference for small viewing angles in the Hβ\beta RM sample and to the substantially smaller values inferred for some AGN relative to the dynamical modeling estimates. A more realistic treatment of cloud irradiation in future versions of the FRADO model may shift the predicted ϕ\phi distribution toward larger values and broaden its range.

Systematic Uncertainties of the Observed AGN Properties

The observed RM MBHM_{\mathrm{BH}} (MobsM_{\mathrm{obs}}) and Eddington ratio (λobs\lambda_{\mathrm{obs}}) deviate from the calibrated mass (MM) and Eddington ratio (λ\lambda) owing to viewing angle effects. By comparing the observed and intrinsic parameters, we discuss the systematic uncertainties of the observed mass and Eddington ratio. Using the same sampling configuration as in § 4.1, we construct joint probability density maps of (MobsM_{\mathrm{obs}}, MM) and (λobs\lambda_{\mathrm{obs}}, λ\lambda) from 4×1044 \times10^{4} paired realizations per AGN. To account for measurement uncertainties, we generate 25 Monte Carlo draws of MobsM_{\mathrm{obs}} and λobs\lambda_{\mathrm{obs}}, select one random draw for each realization, and apply the viewing angle calibration using the sampled θ\theta to obtain MM and λ\lambda.

As presented in Figure 10, the observed MBHM_{\mathrm{BH}} is smaller than MM by 0.49±0.690.49 \pm0.69 dex. In contrast, λobs\lambda_{\mathrm{obs}} is larger than λ\lambda by 0.49±0.720.49 \pm0.72 dex. The upward mass shift suggests that the RM MBHM_{\mathrm{BH}} of the Hβ\beta RM sample may be systematically underestimated, implying that the fraction of super-Eddington AGN may be lower than that estimated using the RM MBHM_{\mathrm{BH}}. In fact, only ∼1.1%\sim1.1\% of the calibrated λ\lambda draws exceed unity, whereas 13 of the 151 AGN (∼8.6%\sim8.6\%) have λobs>1\lambda_{\mathrm{obs}} > 1.

Comparison between observed and intrinsic black hole masses and Eddington ratios

Figure 10. Left: Comparison between the RM MBHM_{\mathrm{BH}} (MobsM_{\mathrm{obs}}) and the intrinsic mass (MM). Right: Comparison between the observed Eddington ratio (λobs\lambda_{\mathrm{obs}}) and the intrinsic λ\lambda. White contours denote the 16%, 50%, and 90% probability regions and the centroids of the distributions are indicated by crosses.

The systematically larger intrinsic MBHM_{\mathrm{BH}} appears to be in tension with S. Wang et al. [55], who reported that dynamical modeling masses are consistent on average with the RM MBHM_{\mathrm{BH}}. By comparing the dynamical-model-based MBHM_{\mathrm{BH}} reported by L. Villafaña et al. [53] with the inferred intrinsic MM for the 31 overlapping AGN, we find an offset of 0.49±0.380.49 \pm0.38 dex. The trend toward larger mass in our Bayesian framework relative to the dynamical modeling estimates is more prominent for high-mass AGN. Since the mass discrepancy is driven by the preference for small inferred viewing angles in the Hβ\beta RM sample, the degeneracy between θ\theta and ϕ\phi may provide an underlying explanation (see § 5.1). In particular, the limited range of ϕ\phi predicted by our model (see also § 5.3) may restrict the corresponding range of θ\theta, leading to smaller inferred viewing angles and hence higher calibrated masses than those obtained from dynamical modeling.

Summary and Conclusions

In this work, we present the updated FRADO model and the Bayesian framework to investigate the viewing angle dependence of the RBLRR_{\mathrm{BLR}}–L5100L_{5100} relation. We summarize our main results as follows:

  • We updated the FRADO model by including a phase factor for anisotropic visibility of the BLR cloud surface to the line-of-sight. We find that the far-side BLR has a larger weight than the near-side BLR in the transfer function, increasing the viewing angle dependence of the BLR time lag predicted by the model.

  • Using a Bayesian framework, we constrained the viewing angle of each AGN in the Hβ\beta RM sample,

finding that the inferred viewing angles are relatively small and broadly consistent with the viewing angle estimates based on dynamical modeling of velocity-resolved lags [53]. Several objects showed significant offsets, presumably due to the large opening angle inferred from dynamical modeling, reflecting the degeneracy between the viewing angle and the BLR opening angle.

  • The updated FRADO combined with a Bayesian framework reproduces the observed RBLRR_{\mathrm{BLR}}–L5100L_{5100} relation. As the viewing angle affects the Hβ\beta time lag and the anisotropic disk luminosity, the slope and scatter of the observed RBLRR_{\mathrm{BLR}}–L5100L_{5100} relation can be reproduced with the inferred viewing angles for the Hβ\beta RM sample.

  • Since the RM MBHM_{\mathrm{BH}} is determined using an average virial factor corresponding to the representative viewing angle of the sample, the FRADO-based MBHM_{\mathrm{BH}} is larger than the RM-based MBHM_{\mathrm{BH}} by ∼0.5\sim0.5 dex because the viewing angles inferred for individual AGN are smaller than the representative viewing angle. This systematic mass offset requires further investigation, particularly regarding the limited range of opening angles predicted by the FRADO model.

Overall, the updated FRADO model successfully reproduces the observed RBLRR_{\mathrm{BLR}}–L5100L_{5100} relation. The dependence of RBLRR_{\mathrm{BLR}}, L5100L_{5100}, and MBHM_{\mathrm{BH}} on viewing angle should be investigated further using both observational and theoretical approaches.

Acknowledgments

This work has been supported by the Basic Science Research Program through the National Research Foundation of the Korean Government (RS-2026-25487266).

References

  1. [1]Amorim, A., Bauböck, M., Brandner, W., et al. 2020, A&A, 643, A154, doi: 10.1051/0004-6361/202039067DOI
  2. [2]Amorim, A., Bauböck, M., Brandner, W., et al. 2021, A&A, 648, A117, doi: 10.1051/0004-6361/202040061DOI
  3. [3]Amorim, A., Bourdarot, G., Brandner, W., et al. 2024, A&A, 684, A167, doi: 10.1051/0004-6361/202348167DOI
  4. [4]Baron, D., & Ménard, B. 2019, MNRAS, 487, 3404, doi: 10.1093/mnras/stz1546DOI
  5. [5]Baskin, A., & Laor, A. 2018, Monthly Notices of the Royal Astronomical Society, 474, 1970, doi: 10.1093/MNRAS/STX2850DOI
  6. [6]Begelman, M. C., & Silk, J. 2017, Monthly Notices of the Royal Astronomical Society, 464, 2311, doi: 10.1093/MNRAS/STW2533DOI
  7. [7]Bentz, M. C., Walsh, J. L., Barth, A. J., et al. 2009, ApJ, 705, 199, doi: 10.1088/0004-637X/705/1/199DOI
  8. [8]Bentz, M. C., Denney, K. D., Grier, C. J., et al. 2013, ApJ, 767, 149, doi: 10.1088/0004-637X/767/2/149DOI
  9. [9]Bian, W., & Zhao, Y. 2004, MNRAS, 352, 823, doi: 10.1111/j.1365-2966.2004.07967.xDOI
  10. [10]Blandford, R. D., & McKee, C. F. 1982, ApJ, 255, 419, doi: 10.1086/159843DOI
  11. [11]Collin, S., Kawaguchi, T., Peterson, B. M., & Vestergaard, M. 2006, A&A, 456, 75, doi: 10.1051/0004-6361:20064878DOI
  12. [12]Czerny, B., & Hryniewicz, K. 2011, A&A, 525, L8, doi: 10.1051/0004-6361/201016025DOI
  13. [13]Czerny, B., Li, Y.-R., Hryniewicz, K., et al. 2017, ApJ, 846, 154, doi: 10.3847/1538-4357/AA8810DOI
  14. [14]Decarli, R., Dotti, M., Fontana, M., & Haardt, F. 2008, MNRAS, 386, L15, doi: 10.1111/j.1745-3933.2008.00451.xDOI
  15. [18]Emmering, R. T., Blandford, R. D., Shlosman, I., et al. 1992, ApJ, 385, 460, doi: 10.1086/170955DOI
  16. [19]Fabian, A. C. 2012, ARA&A, 50, 455, doi: 10.1146/ANNUREV-ASTRO-081811-125521DOI
  17. [20]Ferland, G. J., & Persson, S. E. 1989, The Astrophysical Journal, 347, 656
  18. [21]Grier, C. J., Pancoast, A., Barth, A. J., et al. 2017, ApJ, 849, 146, doi: 10.3847/1538-4357/aa901bDOI
  19. [22]Hardcastle, M. J., & Croston, J. H. 2020, New Astronomy Reviews, 88, 101539, doi: 10.1016/j.newar.2020.101539DOI
  20. [23]Harrison, C. M., & Almeida, C. R. 2024, Galaxies, 12, 17, doi: 10.3390/galaxies12020017DOI
  21. [24]Heckman, T. M., & Best, P. N. 2014, ARA&A, 52, 589, doi: 10.1146/annurev-astro-081913-035722DOI
  22. [25]Hopkins, P. F. 2025, Open Journal of Astrophysics, 8, 56, doi: 10.33232/001c.137969DOI
  23. [26]Jaiswal, V. K., Mandal, A. K., Prince, R., et al. 2025, A&A, 702, A92, doi: 10.1051/0004-6361/202452497DOI
  24. [27]Kaspi, S., Smith, P. S., Netzer, H., et al. 2000, ApJ, 533, 631, doi: 10.1086/308704DOI
  25. [28]Lawrence, A., & Elvis, M. 2010, ApJ, 714, 561, doi: 10.1088/0004-637X/714/1/561DOI
  26. [30]Martínez-Aldama, M. L., Zajaček, M., Czerny, B., & Panda, S. 2020, ApJ, 903, 86, doi: 10.3847/1538-4357/abb6f8
  27. [31]Murayama, T., Taniguchi, Y., & Nagao, T. 2001, in Astronomical Society of the Pacific Conference Series, Vol. 222, Narrow-Line Seyfert 1 Galaxies and Their Place in the Universe, 375
  28. [32]Murray, N., Chiang, J., Grossman, S. A., & Voit, G. M. 1995, ApJ, 451, 498, doi: 10.1086/176238DOI
  29. [33]Naddaf, M. H. 2024, Universe, 11, 8, doi: 10.3390/universe11010008DOI
  30. [34]Naddaf, M. H., Czerny, B., & Szczerba, R. 2020, Front. Astron. Space Sci., 7, 15, doi: 10.3389/fspas.2020.00015DOI
  31. [35]Naddaf, M. H., Czerny, B., & Szczerba, R. 2021, ApJ, 920, 30, doi: 10.3847/1538-4357/ac139dDOI
  32. [36]Naddaf, M. H., Martinez-Aldama, M. L., Hutsemekers, D., Savic, D., & Czerny, B. 2025, A&A, 702, A46, doi: 10.1051/0004-6361/202555767DOI
  33. [37]Nagao, T., Taniguchi, Y., & Murayama, T. 2001, in Astronomical Society of the Pacific Conference Series, Vol. 222, Narrow-Line Seyfert 1 Galaxies and Their Place in the Universe, 379
  34. [38]Netzer, H., & Laor, A. 1993, ApJL, 404, L51, doi: 10.1086/186741
  35. [39]Pan, Z., Jiang, L., Guo, W.-J., et al. 2025, ApJ, 987, 48, doi: 10.3847/1538-4357/add7ddDOI
  36. [40]Pancoast, A., Brewer, B. J., & Treu, T. 2011, ApJ, 730, 139, doi: 10.1088/0004-637X/730/2/139DOI
  37. [41]Pancoast, A., Brewer, B. J., Treu, T., et al. 2014, Monthly Notices of the Royal Astronomical Society, 445, 3073, doi: 10.1093/MNRAS/STU1419DOI
  38. [42]Pancoast, A., Brewer, B. J., Treu, T., et al. 2012, ApJ, 754, 49, doi: 10.1088/0004-637X/754/1/49DOI
  39. [43]Peterson, B. M. 1993, PASP, 105, 247, doi: 10.1086/133140DOI
  40. [44]Peterson, B. M. 2011, https://arxiv.org/pdf/1109.4181
  41. [45]Peterson, B. M. 2014, Space Sci. Rev., 183, 253, doi: 10.1007/s11214-013-9987-4DOI
  42. [46]Peterson, B. M., Ferrarese, L., Gilbert, K. M., et al. 2004, ApJ, 613, 682, doi: 10.1086/423269DOI
  43. [47]Pucha, R., Juneau, S., Mezcua, M., et al. 2026, A New Record Census of Dwarf AGN and a Bimodal M_BH–M_* Scaling Relation with DESI DR1, https://arxiv.org/abs/2606.02699
  44. [48]Rakshit, S., Stalin, C. S., & Kotilainen, J. 2020, ApJS, 249, 17, doi: 10.3847/1538-4357/ab99c5
  45. [50]Shen, Y., Richards, G. T., Strauss, M. A., et al. 2011, ApJS, 194, 45, doi: 10.1088/0067-0049/194/2/45DOI
  46. [51]Sturm, E., Dexter, J., Pfuhl, O., et al. 2018, Nature, 563, 657, doi: 10.1038/s41586-018-0731-9DOI
  47. [52]Vestergaard, M., & Peterson, B. M. 2006, ApJ, 641, 689, doi: 10.1086/500572DOI
  48. [53]Villafaña, L., Treu, T., Wang, S., et al. 2026, ApJ, 1004, 151, doi: 10.3847/1538-4357/ae69cfDOI
  49. [54]Wang, S., & Woo, J. H. 2024, ApJS, 275, 13, doi: 10.3847/1538-4365/AD74F2DOI
  50. [55]Wang, S., Woo, J. H., Villafaña, L., Treu, T., & Gallo, E. 2026, ApJ, 1000, 180, doi: 10.3847/1538-4357/AE4699DOI
  51. [56]Williams, P. R., Pancoast, A., Treu, T., et al. 2018, ApJ, 866, 75, doi: 10.3847/1538-4357/AE0866
  52. [57]Woo, J. H., Kim, J. C., Cho, H., & Wang, S. 2026, ApJ, 1003, 123, doi: 10.3847/1538-4357/ae6109DOI
  53. [58]Woo, J. H., & Urry, C. M. 2002, ApJL, 581, L5
  54. [59]Woo, J. H., Yoon, Y., Park, S., Park, D., & Kim, S. C. 2015, The Astrophysical Journal, 801, 38, doi: 10.1088/0004-637X/801/1/38DOI
  55. [60]Woo, J. H., Wang, S., Rakshit, S., et al. 2024, ApJ, 962, 67, doi: 10.3847/1538-4357/AD132FDOI

Paper details

vlnc:W_zhpmweur
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