Multi-Band Continuum Reverberation Mapping

• Multi-band continuum reverberation mapping is a technique that uses correlated variability across different continuum bands to infer light-travel delays and map the structure of AGN accretion disks.
• It employs multi-wavelength photometric monitoring, enabling the reconstruction of lag–wavelength relations and rigorous tests of standard thin-disk predictions while accounting for BLR contamination and host dilution.
• Recent observational campaigns reveal that while the lag–wavelength trend often follows a τ ∝ λ^(4/3) scaling, the normalization frequently suggests accretion disk sizes that are larger than theoretical expectations.

Multi-band continuum reverberation mapping is a reverberation-based methodology for active galactic nuclei in which correlated variability between multiple continuum bands is used to infer light-travel delays, and hence geometric scales, within the optical/UV continuum-emitting region of the central engine. In its modern form, it generalizes photometric reverberation mapping from line-dominated versus continuum-dominated bands to a broader program of measuring inter-band continuum lags, reconstructing the lag–wavelength relation, and testing accretion-disk structure while accounting for broad-line region contamination, diffuse continuum, and host dilution (Edri et al., 2012, Fausnaugh et al., 2015, Guo et al., 2022).

1. Physical basis and formalism

Continuum reverberation mapping is usually framed within the lamp-post reprocessing picture: a compact central source illuminates a geometrically thin, optically thick disk, and progressively larger radii re-emit at progressively longer wavelengths after a light-travel delay. In a standard Shakura–Sunyaev disk, the effective temperature scales as

$T(R) \propto R^{-3/4},$

which implies

$$R(\lambda) \propto \lambda^{4/3}, \qquad \tau(\lambda) \propto \lambda^{4/3}.$$

A commonly used empirical parameterization is

$$\tau(\lambda) = \alpha\left[\left(\frac{\lambda}{\lambda_0}\right)^{\beta} - 1\right],$$

with the thin-disk expectation $\beta = 4/3$ (Fausnaugh et al., 2015, Jha et al., 2023).

Several campaigns interpret $\alpha$ or an equivalent $R_{\lambda_0}$ as the normalization of the continuum-emitting size scale. In irradiated-disk treatments, the normalization depends on $M$, $\dot M$, $\eta$, and any external heating term, and is often compared against a thin-disk prediction to quantify the now standard disk-size discrepancy (Fausnaugh et al., 2015, Nuñez et al., 2019).

A more general time-domain formulation writes the variable continuum at wavelength $\lambda$ as a convolution of a driver $$R(\lambda) \propto \lambda^{4/3}, \qquad \tau(\lambda) \propto \lambda^{4/3}.$$0 with a wavelength-dependent transfer function: $$R(\lambda) \propto \lambda^{4/3}, \qquad \tau(\lambda) \propto \lambda^{4/3}.$$1 In this language, the measured lag is not a property of a single annulus, but of a responsivity-weighted transfer function whose centroid, peak, and shape depend on wavelength and geometry (Fu et al., 29 Jul 2025). Simulations of thin-disk lamp-post transfer functions show that multi-band light curves are not only shifted in time but also distorted, because the transfer functions are skewed and broaden with wavelength; this “twisted” behavior implies that methods that ignore the transfer-function shape can underestimate accretion-disk sizes by $$R(\lambda) \propto \lambda^{4/3}, \qquad \tau(\lambda) \propto \lambda^{4/3}.$$2, whereas CREAM recovers unbiased sizes in the LSST ugrizy simulations (Chan et al., 2019).

2. Observational architectures and filter design

The observational design of multi-band continuum reverberation mapping is driven by the expected lag amplitude, the need for high S/N light curves, and the requirement to minimize line contamination. A recurrent strategy is to define at least one reference band that is as line-free as possible, then to sample progressively redder continuum windows with either broad-band or custom narrow-band filters (Edri et al., 2012, Fian et al., 2021).

The broad-band photometric reverberation mapping campaign on NGC 4395 is formally a line-lag experiment, but it provides a methodological prototype. Using the Wise Observatory 1m telescope with SDSS $$R(\lambda) \propto \lambda^{4/3}, \qquad \tau(\lambda) \propto \lambda^{4/3}.$$3, $$R(\lambda) \propto \lambda^{4/3}, \qquad \tau(\lambda) \propto \lambda^{4/3}.$$4, and $$R(\lambda) \propto \lambda^{4/3}, \qquad \tau(\lambda) \propto \lambda^{4/3}.$$5, monitored for 9 consecutive nights with about one data point every $$R(\lambda) \propto \lambda^{4/3}, \qquad \tau(\lambda) \propto \lambda^{4/3}.$$6–15 min in each band, the $$R(\lambda) \propto \lambda^{4/3}, \qquad \tau(\lambda) \propto \lambda^{4/3}.$$7 band was taken as pure continuum, $$R(\lambda) \propto \lambda^{4/3}, \qquad \tau(\lambda) \propto \lambda^{4/3}.$$8 as continuum + H$$R(\lambda) \propto \lambda^{4/3}, \qquad \tau(\lambda) \propto \lambda^{4/3}.$$9, and $$\tau(\lambda) = \alpha\left[\left(\frac{\lambda}{\lambda_0}\right)^{\beta} - 1\right],$$0 as essentially continuum-dominated with weak H$$\tau(\lambda) = \alpha\left[\left(\frac{\lambda}{\lambda_0}\right)^{\beta} - 1\right],$$1 contamination (Edri et al., 2012). This same logic underlies continuum RM: establish a clean continuum reference, identify contaminated bands, and use multi-band information to separate components.

Other campaigns use specially designed narrow-band filters to suppress BLR contamination. PG 2130+099 employed four custom narrow-band filters placed in line-free continuum windows in the rest frame, with daily cadence over $$\tau(\lambda) = \alpha\left[\left(\frac{\lambda}{\lambda_0}\right)^{\beta} - 1\right],$$2 months, and detected significant inter-band continuum lags of up to 3 days (Fian et al., 2021). Mrk 509 used narrow bands centered at 4311, 5688, 6208, and 7018 Å with sub-day time sampling and typical flux uncertainties of $$\tau(\lambda) = \alpha\left[\left(\frac{\lambda}{\lambda_0}\right)^{\beta} - 1\right],$$3, allowing lags of up to $$\tau(\lambda) = \alpha\left[\left(\frac{\lambda}{\lambda_0}\right)^{\beta} - 1\right],$$4 days across the optical range to be measured (Nuñez et al., 2019). MCG 08-11-011 extended this narrow-band strategy from $$\tau(\lambda) = \alpha\left[\left(\frac{\lambda}{\lambda_0}\right)^{\beta} - 1\right],$$5 to $$\tau(\lambda) = \alpha\left[\left(\frac{\lambda}{\lambda_0}\right)^{\beta} - 1\right],$$6 Å with daily sampling (Fian et al., 2023).

The AGN STORM campaign on NGC 5548 exemplifies the high-leverage UV–optical architecture: HST continuum windows at 1158, 1367, 1479, and 1746 Å, Swift UVOT bands, and ground-based $$\tau(\lambda) = \alpha\left[\left(\frac{\lambda}{\lambda_0}\right)^{\beta} - 1\right],$$7 plus $$\tau(\lambda) = \alpha\left[\left(\frac{\lambda}{\lambda_0}\right)^{\beta} - 1\right],$$8, giving contiguous coverage from 1158 Å to $$\tau(\lambda) = \alpha\left[\left(\frac{\lambda}{\lambda_0}\right)^{\beta} - 1\right],$$9 Å with cadences from $\beta = 4/3$0 to 1 day (Fausnaugh et al., 2015). Survey-scale implementations then trade wavelength leverage for sample size: ZTF uses $\beta = 4/3$1 at $\beta = 4/3$2-day cadence over multiple years (Guo et al., 2022), while the Dan Zowada Memorial Observatory sample provides $\beta = 4/3$3 over four years with typical 1–3 day cadence (Miller et al., 7 Nov 2025).

Target selection can be decisive. In the IMBH candidate J0249−0815, high-cadence $\beta = 4/3$4 monitoring on 2 m telescopes yielded no significant variability ($\beta = 4/3$5) over 6–10 hr, primarily because the host fraction at 5100 Å was $\beta = 4/3$6–$\beta = 4/3$7; the study argues that future IMBH campaigns should favor targets with higher AGN-to-host flux ratios and use tailored simulations before intensive monitoring (Zuo et al., 2024).

3. Lag extraction, interpolation, and forward modeling

The most widely used lag estimators in multi-band continuum RM are the interpolated cross-correlation function, DRW-based forward models, and transfer-function reconstructions. ICCF remains standard: lags are usually defined as centroids above a threshold such as $\beta = 4/3$8, with uncertainties estimated through FR/RSS Monte Carlo realizations (Fausnaugh et al., 2015, Fian et al., 2021). Variants include zDCF, the von Neumann estimator, MICA, and multivariate correlation approaches (Fian et al., 2021, Fian et al., 2023).

JAVELIN models the driving light curve as a damped random walk and each responding band as a shifted, scaled, and top-hat-smoothed version of that driver. In NGC 5548, ICCF and JAVELIN lags agree well within uncertainties, although JAVELIN errors are systematically smaller, and ICCF values were adopted as conservative (Fausnaugh et al., 2015). PyROA instead constructs a shared driving light curve and infers per-band shifts, scale factors, and smoothing through MCMC; in Mrk 876, PyCCF, JAVELIN, and PyROA gave consistent lag sequences once long-term trends were removed (Miller et al., 2023).

Detrending can be methodologically essential. In Mrk 876, slow variations broadened the cross-correlation function significantly; subtracting a moving boxcar average with a fiducial width of 100 days narrowed the CCFs and stabilized the inter-band lags, which then became consistent across methods (Miller et al., 2023). By contrast, the NGC 4395 photometric RM experiment found that FR alone gave reliable lags with errors $\beta = 4/3$9 hour, whereas FR+RSS tended to overestimate lags by up to $\alpha$0 in a densely sampled light curve (Edri et al., 2012). This suggests that the optimal uncertainty model depends strongly on cadence and window function.

A separate methodological issue is whether the inference engine represents the true transfer function. In simulated LSST-like ugrizy light curves, PyCS and JAVELIN underestimate disk sizes by $\alpha$1, JAVELIN thin-disk fits give sizes $\alpha$2 too small, and only CREAM yields unbiased measurements because it explicitly models the skewed, wavelength-dependent transfer functions of a lamp-post irradiated thin disk (Chan et al., 2019). This result directly affects the interpretation of high-S/N survey light curves, where transfer-function systematics can dominate over statistical errors.

4. BLR contamination, diffuse continuum, and host dilution

A central complication in multi-band continuum reverberation mapping is that broad-band “continuum” filters are often not pure continuum. In NGC 5548, spectral decomposition with LBT/MODS showed that the U and $\alpha$3 bands contain $\alpha$4 Balmer continuum, while $\alpha$5 and $\alpha$6 contain $\alpha$7–$\alpha$8 H$\alpha$9 emission (Fausnaugh et al., 2015). Simulations based on those fractions found upward biases of $R_{\lambda_0}$0–1.2 d in the U/$R_{\lambda_0}$1 lags and $R_{\lambda_0}$2–1.0 d in $R_{\lambda_0}$3, leading to the practical recommendation that U/$R_{\lambda_0}$4 be excluded from disk-size fits in that dataset (Fausnaugh et al., 2015).

The ZTF $R_{\lambda_0}$5 survey extends this contamination problem to ensemble statistics. For 94 type 1 AGN with significant and consistent inter-band lags, the inferred continuum-emitting sizes are larger than the standard thin-disk prediction, the size scales as $R_{\lambda_0}$6 with a scatter of 0.2 dex, and the observed relation can be explained by a photoionization model in which $R_{\lambda_0}$7 of the total flux comes from diffuse BLR emission (Guo et al., 2022). In that interpretation, optical multi-band continuum RM often measures a composite disk + diffuse BLR radius rather than a pure accretion-disk scale.

Direct evidence for this outer continuum component was presented for six AGNs using the ICCF-Cut method. In the Swift U band, an outer component was extracted with lags consistent with the predicted Balmer continuum lag, about half of the H$R_{\lambda_0}$8 lag value, and similar results were obtained with the JAVELIN photometric reverberation mapping model for four of the six objects (Jiang et al., 2024). This establishes that Balmer diffuse continuum can contribute significantly to CRM lags in the rest-frame $R_{\lambda_0}$9 band.

Broad-line contamination can also bias individual red bands. In Mrk 876, the $M$0-band lag shows an excess attributed to variable H$M$1 broad-line emission, and the flux–flux analysis finds both a variable spectrum following $M$2 and an $M$3-band excess that points to strong variable H$M$4 emission in that band (Miller et al., 2023). In MCG 08-11-011, the very steep $M$5 relation is explicitly interpreted as a possible consequence of diffuse continuum emission (Fian et al., 2023).

Host dilution is an additional systematic that can suppress the variability amplitude without directly altering the lag kernel. J0249−0815 is an extreme case: the host contributes $M$6 and $M$7 of the 5100 Å flux in the SDSS and P200 spectra, respectively, and the optical AGN fraction is $M$8, rendering hour-scale continuum RM infeasible at the achieved photometric precision (Zuo et al., 2024). This suggests that target vetting by AGN-to-host contrast is as important as cadence.

5. Empirical results from representative campaigns

The observational literature now spans both intensive single-object campaigns and large survey samples. The table lists representative results that have shaped the field.

Campaign	Bandpass and sampling	Main result
NGC 5548 STORM (Fausnaugh et al., 2015)	1158 Å to $M$9 Å; UV–optical; nearly daily cadence	Lag–wavelength trend broadly consistent with $\dot M$0; best-fit normalization implies a disk radius 3 times larger than the standard thin-disk prediction for $\dot M$1
Mrk 509 (Nuñez et al., 2019)	Custom optical narrow bands; sub-day sampling over two years	Inter-band delays up to $\dot M$2 days; lags consistent with $\dot M$3; disk size factor 1.8 larger than standard thin-disk theory
PG 2130+099 (Fian et al., 2021)	Four line-free narrow bands; daily cadence over six months	Significant lags up to 3 days; derived disk sizes are a factor of 2–6 larger than theoretical disk sizes predicted from the AGN luminosity estimate
MCG 08-11-011 (Fian et al., 2023)	Five narrow bands from $\dot M$4 to $\dot M$5 Å; daily cadence	Lags up to $\dot M$6 days; observed lags larger than standard thin-disk predictions by a factor of $\dot M$7–7; $\dot M$8
Mrk 876 (Miller et al., 2023)	$\dot M$9 over 3.3 years with robotic observatories	Lag of around 13 days from $\eta$0 to $\eta$1; lags roughly follow $\eta$2, but normalization is approximately a factor of 3 longer than expected; $\eta$3-band excess attributed to H$\eta$4
ZTF parent/core samples (Guo et al., 2022)	$\eta$5, $\eta$6-day cadence, 94 AGN parent sample	Continuum-emitting sizes larger than standard thin-disk sizes; $\eta$7 with a scatter of 0.2 dex in the core+local sample

These campaigns are complemented by source-specific results at the low- and high-accretion-rate ends. For the SEAMBH source IRAS 04416+1215, a thin-disk fit with $\eta$8 gives $\eta$9 light days at the $\lambda$0-band rest wavelength, approximately 4 times larger than the Shakura–Sunyaev prediction (Jha et al., 2023). Over a larger baseline, a 4-year optical CRM study of 18 AGN found $\lambda$1, broadly consistent with the $\lambda$2 expectation for continuum reverberation from either the accretion disk or the BLR (Miller et al., 7 Nov 2025).

6. Interpretive tensions, scalability, and emerging applications

A persistent interpretive tension in multi-band continuum RM is that the shape of the lag–wavelength relation is often close to the thin-disk expectation, while the normalization is too large. NGC 5548, Mrk 509, PG 2130+099, Mrk 876, IRAS 04416+1215, and the ZTF sample all exhibit this pattern to varying degrees (Fausnaugh et al., 2015, Nuñez et al., 2019, Fian et al., 2021, Miller et al., 2023, Jha et al., 2023, Guo et al., 2022). One explanatory route emphasizes genuine departures from standard thin-disk assumptions, including vertical structure, irradiation geometry, or super-Eddington effects; another emphasizes BLR diffuse continuum and line contamination; the ensemble evidence suggests that both classes of effects can operate simultaneously.

This ambiguity matters for downstream inferences. In NGC 5548, the optical continuum lags at wavelengths longer than the $\lambda$3 band are equal to or greater than the lags of high-ionization-state emission lines such as He II $\lambda$4 and $\lambda$5, implying that the optical continuum-emitting source has a physical size comparable to the inner BLR (Fausnaugh et al., 2015). The same paper argues that if H$\lambda$6 lags are measured relative to the optical continuum, the true ionizing radius is larger by $\lambda$7 d, so BLR radii can be underestimated when optical rather than UV drivers are used (Fausnaugh et al., 2015). By contrast, the Mrk 509 analysis shows that a larger black hole mass due to the unknown geometry scaling factor $\lambda$8 can reconcile the observed continuum lag normalization with thin-disk theory for that source (Nuñez et al., 2019). This suggests that the disk-size discrepancy is not necessarily identical in origin from object to object.

On scalability, photometric continuum RM is intrinsically survey-compatible. The NGC 4395 broad-band study explicitly noted that LSST will monitor $\lambda$9 quasars over a decade and that broad-band photometric RM can “increase by several orders of magnitude” the number of reverberation-mapped objects (Edri et al., 2012). ZTF has already demonstrated that $$R(\lambda) \propto \lambda^{4/3}, \qquad \tau(\lambda) \propto \lambda^{4/3}.$$00-day cadence $$R(\lambda) \propto \lambda^{4/3}, \qquad \tau(\lambda) \propto \lambda^{4/3}.$$01 light curves can yield significant inter-band lags for a parent sample of 94 AGN (Guo et al., 2022), while the four-year 18 AGN program shows that medium-sized homogeneous optical samples can constrain the global $$R(\lambda) \propto \lambda^{4/3}, \qquad \tau(\lambda) \propto \lambda^{4/3}.$$02–$$R(\lambda) \propto \lambda^{4/3}, \qquad \tau(\lambda) \propto \lambda^{4/3}.$$03 relation (Miller et al., 7 Nov 2025). A plausible implication is that future progress will depend less on merely increasing sample size than on improving decomposition of disk, diffuse BLR continuum, and line contributions.

The method is also being generalized beyond single-SMBH disks. In a model for low-mass-ratio supermassive black hole binaries, the presence of a low-density cavity between an active mini-disk and a circumbinary disk produces a lag–wavelength relation that is flat at short wavelengths and transitions to $$R(\lambda) \propto \lambda^{4/3}, \qquad \tau(\lambda) \propto \lambda^{4/3}.$$04 at long wavelength, rather than following a single $$R(\lambda) \propto \lambda^{4/3}, \qquad \tau(\lambda) \propto \lambda^{4/3}.$$05 law throughout (Fu et al., 29 Jul 2025). Applied to PG 1302−102, this SMBHB model reproduces the inter-band time lags and yields an inferred total mass and orbital period consistent with values derived from other methods (Fu et al., 29 Jul 2025). This extends multi-band continuum RM from a disk-sizing tool to a structural discriminator for more complex accretion flows.

Multi-band continuum reverberation mapping therefore occupies a dual role. At one level it is a precision timing method for measuring $$R(\lambda) \propto \lambda^{4/3}, \qquad \tau(\lambda) \propto \lambda^{4/3}.$$06 and testing $$R(\lambda) \propto \lambda^{4/3}, \qquad \tau(\lambda) \propto \lambda^{4/3}.$$07. At another, it has become a diagnostic of contaminating continua, BLR stratification, host dilution, transfer-function asymmetry, and non-standard accretion geometries. The contemporary literature does not reduce it to a single measurement of a single radius; it treats the continuum lag spectrum as a composite observable whose physical interpretation depends on filter design, cadence, decomposition strategy, and the chosen forward model.