Reverberation Mapping and the Broad-Line Region
You cannot resolve the broad-line region of a quasar with any telescope ever built — a billion-solar-mass black hole gathers its fastest emitting gas into a structure only light-weeks across, subtending microarcseconds on the sky. Yet its size is measured routinely, and the trick is to stop trying to see it and instead to listen for an echo. The accretion disk at the center of an active galactic nucleus flickers stochastically in ultraviolet and optical continuum. That flickering ionizes the surrounding gas, which answers in the broad recombination lines — Hβ, C IV, Mg II — but the answer arrives late, delayed by the extra path light must travel out to the gas and back toward us. Cross-correlate the continuum light curve against the line light curve, read off the lag τ, and the responsivity-weighted radius drops out as R_BLR = cτ. A time delay has become a distance. That distance is only half of a black hole scale. The same broad lines carry a Doppler width ΔV of thousands of kilometres per second, the orbital speedometer of the emitting gas. If those motions are virialized in the black hole's gravity, then R_BLR and ΔV combine into a virial mass M_bullet = f R_BLR ΔV²/G, where the single dimensionless factor f hides everything we do not know about the region's three-dimensional geometry and inclination. Reverberation mapping has been carried out for on the order of a hundred nuclei, anchoring a radius–luminosity relation that lets one spectrum weigh any broad-line quasar to the edge of the observable universe.
- Reverberation mapping turns a time delay into a length: broad emission lines echo the ionizing continuum after a light-travel lag τ, so the broad-line region radius is R_BLR = cτ — light-days to light-weeks, far below any imaging resolution.
- The lag is measured by cross-correlating the continuum and line light curves; its centroid is the responsivity-weighted mean delay, which for a thin shell equals R/c exactly.
- Combining the radius with the emission-line velocity width ΔV gives a virial black hole mass M_bullet = f R_BLR ΔV²/G, with f an order-unity geometry and kinematics factor.
- Photoionization at a characteristic ionization parameter forces a radius–luminosity relation R_BLR ∝ L^(1/2), letting a single spectrum estimate M_bullet across cosmic time.
- The virial factor f is calibrated statistically by requiring reverberation masses to lie on the same M–σ relation as quiescent galaxies; its scatter sets a factor-of-a-few floor on individual masses.
- Key idea: a time delay is a distance — the light-travel lag between continuum flares and the broad-line response measures R_BLR = cτ, and virialized gas motions then convert that size into a black hole mass.
Where this lives today
Reverberation mapping has moved from single-object campaigns to industrial scale. Multi-object programs such as the SDSS Reverberation Mapping project and OzDES monitor hundreds of quasars simultaneously with fibre spectroscopy, pushing lag measurements to redshifts beyond one and to lines like C IV and Mg II that fall in the optical there — the first step toward calibrating single-epoch masses for the flood of spectra from surveys like DESI and, soon, 4MOST. The frontier technique is velocity-resolved reverberation: by measuring the lag as a function of position across the line profile, campaigns now recover velocity-delay maps that distinguish rotating disks from inflows and outflows, turning a size measurement into a partial kinematic map and directly attacking the geometry that the factor f papers over. Continuum reverberation is a parallel industry: inter-band lags between ultraviolet and optical continuum, measured by Swift and ground-based photometry in campaigns like AGN STORM on NGC 5548, size the accretion disk itself and have surfaced a live puzzle — measured disk sizes run larger than the Shakura–Sunyaev thin-disk model predicts. And GRAVITY on the VLT Interferometer has now spatially resolved the broad-line region of a few bright quasars, measuring its size interferometrically and confronting it head-to-head with the reverberation radius, a cross-check that is beginning to pin down f object by object.
The physical picture
Picture the accretion disk as a lamp that flickers at the center of a roughly spherical swarm of gas clouds. When the lamp brightens, the news propagates outward at the speed of light. Clouds on the near side — between us and the source — see the flare first and re-emit their lines soonest; their reprocessed light and the direct continuum reach us almost together, so the lag is nearly zero. Clouds on the far side respond later and their echo must additionally cross the full diameter of the swarm, so they lag by up to 2R/c. Average over the whole geometry and the emission line trails the continuum by exactly the light-crossing time of the radius. The surfaces of constant delay are paraboloids — isodelay surfaces — with the black hole sitting at the focus.
The line width supplies the second half. Gas that orbits faster at a given radius needs more enclosed mass to stay bound, so a broader line means a heavier black hole — the same virial logic that weighs a star cluster, with the crucial simplification that here the enclosed mass is essentially all in the central point. But to weigh anything you must first measure the lag, and the measurement is nothing more than two light curves offset in time. The continuum leads, the line follows, and the offset between their correlated features is the raw datum of the entire method. Everything quantitative in reverberation mapping flows from reading that offset off a pair of time series taken over a monitoring season.
The core equations
The starting point is pure geometry: gas at a given angle around the source contributes a delay set by the extra distance its light travels. Zero the clock on the direct continuum photons; the reprocessed light from a cloud at angle θ then arrives later by the source-to-gas leg plus the projection of the gas position onto the line of sight.
where: τ — observed delay of the line response relative to the continuum; R_BLR — radial distance of the responding gas from the source; θ — angle between the inward line of sight and the gas position vector, so θ = π is the near side (τ = 0) and θ = 0 the far side (τ = 2R/c); c — speed of light. Reading it: the surfaces of constant τ are paraboloids with the black hole at the focus, and the whole region's response is spread over delays from 0 to 2R/c. Averaging equation (1) over an isotropically emitting thin shell collapses that spread to a single number. Because the mean of cos θ over the sphere is zero, the responsivity-weighted centroid lag reduces to the light-crossing time of the radius, which is why the operational radius is simply c times the measured lag.
where: ⟨τ⟩ — responsivity-weighted mean (centroid) lag recovered from cross-correlation. Reading it: this is the definition of the reverberation radius — it is an emission-weighted effective radius, not a sharp edge, and different lines give different R because they respond at different depths. The full description of the response is not a single lag but a distribution over delays, the transfer function Ψ(τ). The line light curve is the continuum light curve convolved with Ψ, so in principle a well-sampled campaign inverts the data for Ψ and maps the geometry directly.
where: L(t) — emission-line light curve; C(t) — driving continuum light curve; Ψ(τ) — transfer function, the delay-weighted response of the line to a continuum impulse. Reading it: the centroid of Ψ is ⟨τ⟩, and its shape encodes inflow, outflow, or Keplerian rotation once resolved in velocity. In practice the lag is extracted by cross-correlation rather than by inverting the convolution. The cross-correlation function of the two light curves peaks at the delay that best aligns their variations, and its centroid is taken as ⟨τ⟩.
where: F_CCF — cross-correlation function of continuum and line; τ — trial lag scanned to find the peak. Reading it: the centroid above some fraction of the peak, not merely the peak itself, is the standard estimator, and its uncertainty is set by bootstrapping the light curves. The radius alone becomes a mass when combined with the line-of-sight velocity width under the assumption of virialized motion. The result is the central formula of the field: a black hole mass built entirely from observables save one dimensionless factor.
where: M_bullet — black hole mass; ΔV — emission-line velocity width, measured as the line dispersion σ_line or the FWHM of the variable spectrum; f — dimensionless virial factor absorbing geometry, kinematics, and inclination; G — gravitational constant. Reading it: f is order unity, calibrated to ⟨f⟩ ≈ 4–5 when ΔV = σ_line, and it is the dominant systematic in any single mass. Because photoionized gas responds most strongly where the ionizing flux produces a characteristic ionization state, holding the ionization parameter fixed ties the responding radius to the source luminosity. The empirical slope confirms the prediction.
where: L_5100 — monochromatic AGN continuum luminosity at 5100 Å; α — empirical slope, consistent with the photoionization value 1/2. Reading it: this relation, calibrated on reverberation-mapped nuclei, is what lets a single spectrum stand in for a whole monitoring campaign. The photoionization argument itself rests on the ionization parameter, the ratio of ionizing photon density to gas density at the cloud.
where: U — dimensionless ionization parameter; Q(H) — rate of hydrogen-ionizing photons from the source; n_H — hydrogen number density of the BLR gas; R — distance of the gas from the source. Reading it: fixing U and n_H forces R² ∝ Q(H) ∝ L, the seed of the radius–luminosity relation derived below.
The goal is to chain two independent measurements — a light-travel time lag and an emission-line width — into one black hole mass that depends only on observables plus a single dimensionless calibration factor. We assume the broad-line gas is gravitationally bound to and dominated by the central point mass, that the very same gas which produces the lag also produces the width, and that its motions are virialized rather than dominated by radiation pressure or winds. From these three assumptions we assemble M_bullet = f c⟨τ⟩ΔV²/G and check that it carries units of mass.
Step 1. Extract the lag as the centroid of the cross-correlation function, ⟨τ⟩ = centroid[F_CCF(τ)], scanning trial delays until the shifted line light curve best matches the continuum. Why: the centroid, not the bare peak, is the responsivity-weighted mean delay of the transfer function, which is the quantity the shell average in Step 2 converts cleanly to a radius.
Step 2. Convert the lag to a radius with R_BLR = c⟨τ⟩, justified because the isodelay relation τ = (R/c)(1 + cos θ) averages over an isotropic shell to ⟨τ⟩ = R/c. Why: this is the whole conceptual leap of the method — the temporal measurement becomes a spatial one with no free parameters, since the speed of light is exact.
Step 3. Impose the virial condition for gas bound to and dominated by the point mass, ⟨v²⟩ = G M_bullet/R_BLR, the mean-square speed of a Keplerian or virialized ensemble around a central mass. Why: only if the black hole's gravity dominates over gas self-gravity, disk mass, and radiation pressure does the enclosed mass reduce to M_bullet alone, which is what makes the region a clean scale.
Step 4. Relate the true speed to the observed line width, ⟨v²⟩ = g ΔV², where g is a projection and geometry constant linking the line-of-sight width ΔV to the three-dimensional speed. Why: we only ever observe one velocity component through the Doppler width, so an unknown factor connecting ΔV to ⟨v²⟩ is unavoidable and must be folded into the calibration.
Step 5. Solve for the mass, combining the virial coefficient and g into the single factor f, giving M_bullet = f R_BLR ΔV²/G = f c⟨τ⟩ΔV²/G. Dimensionally [c τ V²/G] = (m)(m² s⁻²)/(m³ kg⁻¹ s⁻²) = kg, a mass, and f ≈ 4–5 for ΔV = σ_line. Why: every quantity except f is measured, so the accuracy of an individual mass is limited entirely by how well f — the region's unknown geometry — is known, which is calibrated statistically rather than per object.
Go deeper · Derivation B · 2 minThe thin-shell lag average and the isodelay paraboloid
The goal is to prove the identity R_BLR = c⟨τ⟩ that underlies every reverberation radius, showing explicitly that the light-echo delays of a spherical shell, though spread continuously from zero on the near side to twice the light-crossing time on the far side, average to exactly the light-crossing time of the radius. We work from the extra path length of reprocessed light, weight by solid angle for isotropic emission, and evaluate the resulting integral, so that the deceptively simple R = cτ is placed on a firm geometric footing.
Step 1. Compute the extra path for a cloud at angle θ. The reprocessed light travels the source-to-gas leg R_BLR and then reaches the observer along a path shorter or longer than the direct one by R_BLR cos θ, so the total excess is ℓ = R_BLR(1 + cos θ) and the delay is τ(θ) = (R_BLR/c)(1 + cos θ). Why: this is the entire physical content — the delay is set by geometry alone, independent of the gas physics, provided the gas responds fast compared with the variability.
Step 2. Recognize the isodelay surfaces. Setting τ constant fixes 1 + cos θ, which in three dimensions is a paraboloid of revolution with the source at its focus, opening away from the observer. Why: identifying the constant-delay surfaces as paraboloids explains why the response smears out in time and why velocity-resolved lags can map the geometry.
Step 3. Average over an isotropically emitting thin shell, weighting each ring by its solid angle sin θ dθ, so ⟨τ⟩ is the ratio of ∫₀^π (R/c)(1 + cos θ) sin θ dθ to ∫₀^π sin θ dθ. Why: the responsivity-weighted centroid the cross-correlation returns is precisely this solid-angle average, not the peak or the near-side minimum, so the correct estimator is the mean.
Step 4. Evaluate. The numerator integral is ∫₀^π (1 + cos θ) sin θ dθ = 2 because the cos θ term integrates to zero over the sphere, and the denominator is also 2, so ⟨τ⟩ = R_BLR/c and therefore R_BLR = c⟨τ⟩. Why: the vanishing of the cos θ term is what makes the effective radius come out clean — the near-side head start and far-side lag cancel in the mean, leaving the bare light-crossing radius.
Go deeper · Derivation C · 1 minPhotoionization origin of the radius-luminosity relation
The goal is to show that a single physical assumption — that the broad lines are emitted preferentially by gas held at a characteristic ionization parameter and density — forces the responding radius to scale as the square root of the AGN luminosity, R_BLR ∝ L^(1/2), reproducing the empirically measured slope near 0.5. This is what elevates reverberation mapping from a per-object curiosity into a survey tool, because it lets the luminosity of a single spectrum stand in for a lag that was never measured.
Step 1. Write the ionization parameter U = Q(H)/(4π R² c n_H), the ratio of the ionizing photon number density at radius R to the gas density n_H. Why: photoionization models show the emissivity of a given line peaks at a fairly narrow range of U, so the strongest response comes from gas near a preferred U rather than from all radii equally.
Step 2. Note that the ionizing photon rate scales with luminosity, Q(H) ∝ L, for a roughly universal ionizing spectral shape across AGN. Why: without an approximately fixed spectral energy distribution the mapping from luminosity to photon rate would itself carry scatter, so this near-universality is what keeps the relation tight.
Step 3. Hold U and n_H fixed at their characteristic emitting values, so that R² ∝ Q(H)/(U n_H) ∝ L. Why: if the line-emitting gas always sits where U takes its preferred value, then a brighter nucleus must push that surface outward to keep the ionizing flux — and hence U — the same.
Step 4. Take the square root, R_BLR ∝ L^(1/2), matching the calibrated slope α ≈ 0.5 (Bentz et al. 2013). Why: agreement between this first-principles slope and the measured one confirms that photoionization, not some coincidence, controls the size of the broad-line region, which is why the relation extrapolates reliably to luminosities and redshifts never monitored.
Limiting cases
The reverberation formalism has clean, checkable behavior in identifiable regimes:
| Case | Result | Meaning |
|---|---|---|
| Instantaneous gas response | t_rec ≪ τ_var | The recombination time is short compared with continuum variability, so the line tracks the ionizing flux and the observed delay is pure light-travel: R = cτ holds cleanly. |
| Black-hole-dominated dynamics | M_bullet ≫ M_gas, M_disk( | The point mass dominates the potential, so the enclosed mass in the virial relation is the black hole alone and the mass estimate is valid. |
| Radius–luminosity scaling | R_BLR ∝ L^(1/2) | Photoionization at fixed ionization parameter and density fixes the emitting radius, giving the empirical R–L relation used for single-epoch masses. |
| Well-sampled light curves | Δt_samp ≲ τ ≲ T_camp | The lag is recoverable only if the cadence resolves it and the campaign is long enough to contain several independent variations. |
| Face-on flattened BLR | ΔV → small, M underestimated | A rotating disk of clouds seen pole-on shows a narrow line at a real radius, so the naive virial mass is too low — exactly the degeneracy the factor f absorbs. |
Worked examples
1 · A Seyfert black hole from a monitoring campaign (fully worked)
A reverberation campaign on a Seyfert nucleus measures an Hβ lag of 20 light-days from the cross-correlation centroid and a line dispersion σ_line = 2000 km/s in the rms spectrum. Take the ensemble virial factor f = 4.3. What is the black hole mass?
Step 1. Convert the lag to a radius. One light-day is c × 86 400 s = 2.59×10¹³ m, so R_BLR = cτ = 20 × 2.59×10¹³ m = 5.18×10¹⁴ m (about 35 light-hours of black hole gravity, or ~0.017 pc). This is the responsivity-weighted radius of the Hβ-emitting shell.
Step 2. Form the virial product. With σ_line = 2000 km/s = 2.0×10⁶ m/s, the product is R_BLR σ²/G = (5.18×10¹⁴)(2.0×10⁶)²/(6.674×10⁻¹¹) = 3.10×10³⁷ kg. Dividing by M_sun = 1.989×10³⁰ kg gives a virial product of 1.56×10⁷ M_sun. This is the mass up to the factor f — the model-independent measured quantity.
Step 3. Apply the virial factor. M_bullet = f × 1.56×10⁷ M_sun = 4.3 × 1.56×10⁷ = 6.7×10⁷ M_sun. A textbook Seyfert-galaxy black hole, weighed from nothing but two light curves and a line width.
2 · A single-epoch mass from one quasar spectrum
A quasar spectrum gives a continuum luminosity L_5100 = 10⁴⁴ erg/s and a broad Hβ FWHM of 4000 km/s. Estimate R_BLR from the radius–luminosity relation and hence the black hole mass, using the FWHM virial normalization f ≈ 1.
Step 1. Apply the calibrated R–L relation log₁₀(R_BLR/light-day) = 1.527 + 0.533 log₁₀(L_5100/10⁴⁴ erg s⁻¹). At L_5100 = 10⁴⁴ the second term vanishes, so R_BLR = 10^1.527 = 33.7 light-days = 8.7×10¹⁴ m. One luminosity has replaced a whole monitoring campaign.
Step 2. Insert into the virial formula with ΔV = FWHM = 4000 km/s = 4.0×10⁶ m/s and f ≈ 1: M_bullet = (8.7×10¹⁴)(4.0×10⁶)²/(6.674×10⁻¹¹) = 2.1×10³⁸ kg. Note that f differs from Example 1 because FWHM, not σ_line, is the width estimator here.
Step 3. Convert to solar masses: M_bullet = 2.1×10³⁸ / 1.989×10³⁰ ≈ 1.0×10⁸ M_sun. A hundred-million-solar-mass black hole from a single spectrum — the workhorse estimate behind quasar demographics across cosmic time.
3 · Why the region cannot be imaged
The Seyfert of Example 1 lies at 30 Mpc. What angular size does its 5.18×10¹⁴ m broad-line region subtend, and how does that compare with the sharpest optical resolution?
Step 1. The angular size is θ ≈ R/d = 5.18×10¹⁴ m / (30 × 3.086×10²² m) = 5.6×10⁻¹⁰ rad. Converting, 5.6×10⁻¹⁰ rad × 206 265 arcsec/rad ≈ 1.2×10⁻⁴ arcsec = 120 microarcsec. This is the radius; the diameter is a couple of hundred microarcseconds.
Step 2. The diffraction limit of an 8 m optical telescope is roughly λ/D ≈ 5×10⁻⁷/8 ≈ 6×10⁻⁸ rad ≈ 13 milliarcsec — about a hundred times too coarse. Direct imaging is hopeless by two orders of magnitude, which is exactly why the time-domain echo is the only route to R_BLR for all but the very nearest nuclei.
Physics of the portrait
Reverberation mapping is a rare instrument built almost entirely from a single idea — that a delay is a distance — pushed relentlessly against the practical difficulty of monitoring faint, slowly varying nuclei for years at a time.
Origins · 1 minreverberation mapping and how this was discovered
The seed of the method predates the name. As early as the 1970s observers noticed that the broad emission lines of Seyfert galaxies varied in step with the continuum but delayed, and interpreted the delay as a light-travel time across the line-emitting region. The formalism was made rigorous in 1982 by Roger Blandford and Christopher McKee, who cast the line response as a convolution of the continuum with a transfer function Ψ(τ) and coined the term reverberation mapping — the ambition, never fully realized in practice, of inverting the light curves for the full velocity-delay map of the gas. Through the late 1980s and 1990s the observational program was carried forward above all by Bradley Peterson and the International AGN Watch, whose coordinated campaigns on NGC 5548 and a handful of other Seyferts turned the idea into repeatable lag measurements and established that the broad-line region is stratified, with higher-ionization lines responding at smaller radii. The decisive conceptual bridge came in 2000, when the discovery that quiescent galaxies obey a tight M–σ relation gave reverberation masses an external calibration: requiring active galaxies to share that relation fixes the ensemble virial factor f, a step carried out by Onken and collaborators in 2006. Bentz and collaborators then cleaned the radius–luminosity relation of host-galaxy starlight in 2013, tightening it enough to anchor single-epoch masses for hundreds of thousands of quasars. What began as a puzzling delay in a Seyfert spectrum is now the primary black hole scale for the entire active universe.
Going further
Texts. Carroll and Ostlie, An Introduction to Modern Astrophysics, 2nd ed., Ch. 28 (Active Galaxies) · Peterson, An Introduction to Active Galactic Nuclei (1997), the standard monograph · Mo, van den Bosch and White, Galaxy Formation and Evolution, Ch. 14 (AGN and black holes) · Binney and Tremaine, Galactic Dynamics, 2nd ed. (virial theorem foundations) · Blandford and McKee, ApJ 255, 419 (1982); Peterson et al., ApJ 613, 682 (2004); Bentz et al., ApJ 767, 149 (2013). The key numerical checks are: the thin-shell average ⟨τ⟩ = R/c follows from ∫₀^π (1 + cos θ) sin θ dθ = 2; the dimensions of M_bullet = f c τ ΔV²/G reduce cleanly to kilograms; the R–L slope of 1/2 comes from the constant-U photoionization argument; and the worked examples give Example 1 (τ = 20 light-days, σ = 2000 km/s, f = 4.3 → 6.7×10⁷ M_sun) and Example 2 (L_5100 = 10⁴⁴ erg/s → R = 33.7 light-days, FWHM = 4000 km/s, f = 1 → ~10⁸ M_sun). The virial factor f ≈ 4–5 for σ_line is the ensemble value calibrated against the M–σ relation (Onken et al. 2004; Woo et al. 2015), and the R–L coefficients 1.527 and 0.533 are from the Bentz et al. 2013 clean-sample fit.
Prerequisites. Active Galactic Nuclei: Unified Model · Supermassive Black Holes · Eddington Luminosity and Accretion Limits · Thin Accretion Disks and the Shakura–Sunyaev Model. Leads to. M–σ Relation and Black Hole–Galaxy Coevolution · Quasars and Quasar Evolution · Iron Kα Line and Relativistic Reflection.
Ask the tutor: Ask me to walk through a reverberation mass from real NGC 5548 light curves, or to show how velocity-resolved lags tell inflow from a rotating disk.
Why is the delay a light-travel time rather than a physical response time of the gas?
Because the gas responds almost instantly on the relevant scale. In the dense broad-line region the recombination time — how long the gas takes to re-establish ionization equilibrium after the continuum changes — is minutes to hours, while the continuum varies on timescales of days to weeks. The gas therefore tracks the ionizing flux essentially in real time in its own frame, and the only meaningful delay left is geometric: the extra distance the light travels from the source out to the cloud and then to us. That is why the measured lag reads out a distance, R = cτ, and not some chemistry of the gas.
What is the virial factor f, and why can we not just calculate it?
The factor f converts the observed line width into a true space velocity and the reverberation radius into a virial radius. It bundles everything we do not know about the region: whether the clouds form a sphere or a flattened disk, whether they rotate, inflow, or outflow, and at what angle we view them. A face-on rotating disk, for instance, shows a narrow line even at a real radius, so its naive mass is too low and f must be large to compensate. Because we cannot image the geometry, f cannot be computed per object; it is calibrated statistically by demanding that reverberation-mapped active galaxies fall on the same M–σ relation as quiescent ones, which fixes the ensemble average near 4–5 for the σ_line estimator.
Why must the line width come from the rms spectrum rather than the mean spectrum?
The lag is produced only by the gas that actually varies in response to the continuum. The rms spectrum, formed from the scatter of each pixel about its time average, isolates precisely that variable component and suppresses any constant emission. Measuring the width there guarantees you are characterizing the same responding gas whose light-travel time set the lag, so the radius and the velocity refer to one physical region. The mean spectrum, by contrast, includes non-varying gas — a narrow-line component, or broad gas at other radii that happens not to respond — whose velocity does not reflect the reverberating shell and biases the mass, sometimes badly.
How does one spectrum estimate a mass when reverberation needs a whole campaign?
Through the radius–luminosity relation. Reverberation campaigns on roughly a hundred nuclei established that R_BLR scales as the square root of the continuum luminosity, a relation grounded in photoionization physics. So for a quasar you never monitored, you measure its continuum luminosity from a single spectrum, read R_BLR off the calibrated relation, take the broad-line width from that same spectrum, and plug both into the virial formula. These single-epoch masses inherit the scatter of the R–L relation and the f uncertainty, so they are good to a factor of a few — but they can be applied to millions of spectra, which is why they underpin nearly all quasar demographics.
Can reverberation mapping do more than measure a size — can it map the gas motions?
Yes, and that is the current frontier. Instead of collapsing the line to a single light curve, velocity-resolved reverberation measures the lag separately in each velocity bin across the line profile. The pattern of lag versus velocity distinguishes the kinematics: symmetric lags point to virialized rotation, red wings that lag less than blue indicate infall, and the reverse signals outflow. Pushed further, the full velocity-delay map recovers the two-dimensional transfer function Ψ(τ, v), a partial tomogram of the region's geometry and dynamics. This is exactly the information that constrains f directly, turning the method from a bulk size measurement into a genuine, if coarse, map of gas within light-days of a black hole.
Why are accretion-disk sizes from continuum reverberation a puzzle?
Inter-band continuum lags — bluer bands leading redder ones by up to a day or so — measure the light-crossing size of the accretion disk, because different wavelengths peak at different disk radii and temperatures. The Shakura–Sunyaev thin-disk model predicts how those lags should grow with wavelength, and campaigns such as AGN STORM on NGC 5548 confirmed the expected wavelength scaling. The trouble is the normalization: the measured disks come out systematically larger, by factors of a few, than the standard model predicts for the known accretion rate. Whether the fix is a revised temperature profile, disk winds, diffuse continuum from the broad-line region contaminating the lags, or scattering remains an open question in accretion physics.