Photometric Reverberation Mapping (PRM)

• Photometric Reverberation Mapping (PRM) is the photometric analogue of spectroscopic reverberation mapping, using time-series imaging to isolate delayed AGN continuum responses.
• The method employs strategic filter pairings—one continuum-dominated and one line-rich—to extract time delays and estimate black hole masses via lag measurements.
• PRM techniques extend to broadband, narrow, medium-band, and polarimetric applications, enabling population-scale studies and diverse analyses of AGN substructures.

Photometric reverberation mapping (PRM) is the photometric analogue of spectroscopic reverberation mapping in which time-series imaging, rather than multi-epoch spectroscopy, is used to recover delayed responses to AGN continuum variability. In its canonical broad-line-region (BLR) form, one band is selected to be continuum-dominated and another to contain continuum plus a broad emission line; the measured delay yields $R_{\rm BLR}=c\tau$, and, with an independently measured line width, enters the virial black-hole-mass estimator. The same general logic has also been extended to continuum reprocessing lags across the accretion disk, to lensed quasars with microlensing-induced contrast changes, and to polarimetric observables, so PRM is best understood as a family of photometric time-delay methods rather than a single filter prescription (Chelouche et al., 2011, Edri et al., 2012, Nuñez et al., 2019, Sluse et al., 2014, Shablovinskaya et al., 2023).

1. Physical basis and formal definitions

The physical basis of PRM is the same as in classical reverberation mapping: continuum variations from the accretion flow or central ionizing source drive a delayed response in reprocessing regions. For BLR work, the characteristic size follows from

$R_{\rm BLR} \equiv c\tau,$

and, if the gas is treated as virialized, the mass estimate is written as

$M_{\rm BH}=f\,\frac{R_{\rm BLR}\,\Delta V^2}{G},$

or equivalently in closely related notations with $V$, $T$, and $C$ for line width, lag, and the speed of light (Haas et al., 2011, Edri et al., 2012, Bachev et al., 2014).

In broadband PRM, the essential decomposition is that a line-rich band contains an instantaneous continuum term plus a delayed line term. A standard representation is

$f_y(t)=f(t)+f_l(t),$

with $f(t)$ the continuum and $f_l(t)$ the line component. The BLR response itself is frequently written as a convolution,

$$f_{\rm BLR}(t)=\int d\tau\,\Psi(\tau)\,f_{\rm cont}(t-\tau),$$

where $R_{\rm BLR} \equiv c\tau,$0 is the transfer function. Spectroscopic reverberation mapping separates line and continuum directly in wavelength space; PRM instead separates them statistically in variability space by exploiting the fact that one band is line-poor and another is line-rich (Edri et al., 2012, 1711.02167, Chelouche et al., 2011).

A central estimator in broad-band PRM is the difference between the cross-correlation of a continuum band and a line-contaminated band and the autocorrelation of the continuum band,

$R_{\rm BLR} \equiv c\tau,$1

The logic is that subtracting the continuum autocorrelation suppresses the zero-lag continuum similarity and isolates the delayed line response. Closely related estimators were introduced in early quasar work as

$R_{\rm BLR} \equiv c\tau,$2

and

$R_{\rm BLR} \equiv c\tau,$3

again using the contrast between a line-rich and line-poor band to expose the delayed component (Edri et al., 2012, Chelouche et al., 2011, Jankov et al., 2021).

2. Observational architectures and filter design

PRM has been implemented with broad-band, narrow-band, and medium-band photometry, with the choice driven by redshift, line equivalent width, throughput, and cadence requirements.

Approach	Band strategy	Representative implementations
Broadband BLR PRM	Continuum-dominated band plus broad band containing continuum + line	NGC 4395 $R_{\rm BLR} \equiv c\tau,$4 (Edri et al., 2012), Mrk 279 $R_{\rm BLR} \equiv c\tau,$5 (Bachev et al., 2014), four Seyfert 1 galaxies with $R_{\rm BLR} \equiv c\tau,$6, $R_{\rm BLR} \equiv c\tau,$7, or $R_{\rm BLR} \equiv c\tau,$8 (Ma et al., 2023)
Narrow/medium-band BLR PRM	One filter on the broad line and one on adjacent continuum	VYSOS-6 H$R_{\rm BLR} \equiv c\tau,$9/H$M_{\rm BH}=f\,\frac{R_{\rm BLR}\,\Delta V^2}{G},$0 monitoring (Haas et al., 2011, Nuñez et al., 2013), Zeiss-1000 SED filters of width $M_{\rm BH}=f\,\frac{R_{\rm BLR}\,\Delta V^2}{G},$1 (Uklein et al., 2019, Malygin et al., 2019)
Continuum PRM	Multiple line-free continuum bands across optical/UV wavelengths	Mrk509 narrow-band continuum campaign (Nuñez et al., 2019), LSST accretion-disk simulations (Nuñez et al., 2022)
Contrast-assisted variants	Single-band or filter-paired decomposition aided by lensing or polarization	Microlensing-aided PRM in lensed quasars (Sluse et al., 2014), medium-band polarimetric reverberation mapping (Shablovinskaya et al., 2023)

Filter placement is the decisive observational design variable. In NGC 4395, the $M_{\rm BH}=f\,\frac{R_{\rm BLR}\,\Delta V^2}{G},$2 band contains broad H$M_{\rm BH}=f\,\frac{R_{\rm BLR}\,\Delta V^2}{G},$3 plus continuum, the $M_{\rm BH}=f\,\frac{R_{\rm BLR}\,\Delta V^2}{G},$4 band contains H$M_{\rm BH}=f\,\frac{R_{\rm BLR}\,\Delta V^2}{G},$5 plus continuum, and the $M_{\rm BH}=f\,\frac{R_{\rm BLR}\,\Delta V^2}{G},$6 band is treated as pure continuum; the variable broad-line fractions are modest but non-negligible, with H$M_{\rm BH}=f\,\frac{R_{\rm BLR}\,\Delta V^2}{G},$7 contributing about 7% of the flux in $M_{\rm BH}=f\,\frac{R_{\rm BLR}\,\Delta V^2}{G},$8 and H$M_{\rm BH}=f\,\frac{R_{\rm BLR}\,\Delta V^2}{G},$9 about 3% in $V$0 (Edri et al., 2012). In the four-Seyfert broadband study, H$V$1 contributes only about 10–30% of the line band, while H$V$2 contamination in the continuum band is usually $V$3, which makes broad-band line PRM substantially harder than narrow-band work (Ma et al., 2023).

Narrow- and medium-band implementations trade throughput for cleaner decomposition. Early small-telescope campaigns used a broad continuum band together with a narrow band centered on H$V$4 or H$V$5, so that a synthetic emission-line light curve could be built by subtracting a scaled continuum light curve (Haas et al., 2011, Nuñez et al., 2013). The Zeiss-1000 program generalized this to medium-band interference SED filters with bandwidth $V$6, covering $V$7–$V$8 in $V$9 steps, allowing object-by-object selection of a line filter and a nearby continuum filter over $T$0 (Malygin et al., 2019, Uklein et al., 2019).

Continuum PRM uses the same logic but targets disk reprocessing instead of the BLR. The Mrk509 campaign employed specially designed narrow-band filters at $T$1 Å, $T$2 Å, $T$3 Å, and $T$4 Å to minimize line and pseudo-continuum contamination while retaining accurate flux calibration over a large field of view (Nuñez et al., 2019). For LSST-like work, band choice becomes redshift-dependent: at very low redshift, $T$5 can contain H$T$6, $T$7 H$T$8, and $T$9 serve as continuum, whereas at higher redshift broad UV lines such as Mg II, C III], and C IV move into the accessible bands (Jankov et al., 2021).

3. Lag extraction, flux decomposition, and inference machinery

The core technical problem in PRM is to recover a delayed component that is diluted by continuum, host-galaxy light, and, in broad-band data, often by additional lines. The earliest line-extraction recipe was explicit continuum subtraction in flux space,

$C$0

with the continuum scale factor estimated from the spectrum or from the relative band contributions. In Ark120, for example, the synthetic line curve was constructed as $C$1; in 3C120 the synthetic H$C$2 curve was similarly defined as $C$3 (Haas et al., 2011, Nuñez et al., 2013).

Broadband H$C$4 work later reformulated this strategy as ICCF-Cut, in which the extracted line light curve is

$C$5

or, in the conservative minimum-ratio form used for NGC 4395 broad-band data,

$C$6

The method is designed to avoid over-subtraction when the spectral estimate of the line fraction is uncertain or non-contemporaneous (Ma et al., 2023, Gu et al., 2024).

For uneven sampling, the interpolated cross-correlation function (ICCF) remains widely used. In the Wise Observatory NGC 4395 campaign, the local ICCF variant was used so that correlations were computed only over overlapping regions and interpolation across nightly gaps was minimized. The lag workflow was: compute the CCF, compute the ACF for the continuum band, form $C$7, identify the maximum, keep the region above 80% of the peak, and compute a weighted centroid lag (Edri et al., 2012). Other programs used the DCF, the Z-transformed DCF, or centroid definitions such as $C$8 in ICCF/DCF analyses of continuum lags (Nuñez et al., 2019, Jankov et al., 2021, Nuñez et al., 2022).

Model-based inference has become equally important. JAVELIN treats the continuum as a damped random walk (DRW) and the responding light curve as a shifted, smoothed, scaled version of that continuum, usually via a top-hat transfer function for BLR applications. In broadband line work, the Pmap and DPmap variants were used to handle mixed continuum-plus-line bands and, when required, small inter-continuum delays (Ma et al., 2023, Gu et al., 2024). Medium-band monitoring on the Zeiss-1000 telescope used JAVELIN with at least 10,000 MCMC samples to obtain lag posteriors for two demonstration objects (Malygin et al., 2019).

Uncertainty estimation is not uniform across the literature. FR/RSS is standard in many campaigns, including the Mrk509 continuum study with 2000 realizations and the broad-band NGC 4395 ICCF-Cut analysis with 1000 realizations (Nuñez et al., 2019, Gu et al., 2024). The Wise NGC 4395 study, however, argued that FR alone was more appropriate than FR/RSS for highly regular intranight cadence, since RSS could overinflate the lag or destroy the signal (Edri et al., 2012). Preprocessing is also central: host subtraction has been handled with the classical flux variation gradient (FVG) method and, more recently, with the probabilistic FVG (PFVG), which was reported to recover host-galaxy fluxes to within 1% precision as long as the light curves do not show a significant contribution from time delays (Gianniotis et al., 2021).

4. Representative empirical results

Early demonstrations established that PRM could reproduce spectroscopic-scale lags in favorable low-redshift systems. The Wise Observatory broad-band campaign on NGC 4395 obtained over 250 data points per filter over 9 consecutive nights and measured an H$C$9 lag of $f_y(t)=f(t)+f_l(t),$0 hours from $f_y(t)=f(t)+f_l(t),$1- and $f_y(t)=f(t)+f_l(t),$2-based combinations, yielding $f_y(t)=f(t)+f_l(t),$3; the lag was described as comparable to previous spectroscopic reverberation results (Edri et al., 2012). A later broad-band NGC 4395 analysis using minute-cadence FTN and GTC light curves, together with ICCF-Cut, JAVELIN, and $f_y(t)=f(t)+f_l(t),$4 methods, recovered an H$f_y(t)=f(t)+f_l(t),$5 lag of approximately $f_y(t)=f(t)+f_l(t),$6–$f_y(t)=f(t)+f_l(t),$7 minutes and obtained $f_y(t)=f(t)+f_l(t),$8, again in agreement with narrow-band PRM and spectroscopic work (Gu et al., 2024).

Broad-band PRM was also demonstrated in longer-lag Seyfert systems. In Mrk 279, standard broad-band $f_y(t)=f(t)+f_l(t),$9 photometry isolated broad H$f(t)$0 variability and gave a lag of approximately 8 days, implying $f(t)$1 light days and $f(t)$2; the result was described as rather consistent with earlier spectroscopic values of $f(t)$3 and $f(t)$4 days (Bachev et al., 2014). In four Seyfert 1 galaxies—MCG+08-11-011, NGC 2617, 3C 120, and NGC 5548—broadband ICCF-Cut, JAVELIN, and $f(t)$5 analyses recovered H$f(t)$6 lags in a range from 9 to 19 days, and the agreement among the three methods, together with comparison to contemporaneous spectroscopic H$f(t)$7 lags, was used as the reliability criterion (Ma et al., 2023).

Narrow- and medium-band campaigns provided high-S/N proof-of-concept benchmarks with small telescopes. In PG0003+199, where H$f(t)$8 contributed about 85% of the narrow-band flux, the rest-frame lag was $f(t)$9 days and the virial mass estimate was $f_l(t)$0. In Ark120, where H$f_l(t)$1 contributed only 50% of the narrow-band flux, the synthetic-line method recovered a rest-frame lag of $f_l(t)$2 days and $f_l(t)$3 (Haas et al., 2011). For 3C120, a five-month campaign with a median sampling of two days measured a rest-frame H$f_l(t)$4 lag of $f_l(t)$5 days and derived $f_l(t)$6 (Nuñez et al., 2013). The Zeiss-1000 medium-band program reported preliminary lags for 2MASX J08535955+7700543 of $f_l(t)$7 days and $f_l(t)$8 days, consistent with each other and with radius–luminosity expectations (Uklein et al., 2019).

PRM has also been used to measure continuum rather than line delays. In Mrk509, two years of optical continuum monitoring with specially designed narrow-band filters achieved sub-day time sampling and typical flux uncertainties of 1%, yielding inter-band continuum delays of up to $f_l(t)$9 days across the optical range. The delay spectrum was consistent with $$f_{\rm BLR}(t)=\int d\tau\,\Psi(\tau)\,f_{\rm cont}(t-\tau),$$0, but the inferred disk size was a factor of 1.8 larger than standard thin-disk predictions (Nuñez et al., 2019).

5. Systematics, assumptions, and recurring methodological disputes

The most persistent limitation in BLR PRM is line dilution. Broadband filters are continuum-dominated by construction, so the reverberation signal can be only a few percent of the measured flux. Tests on SDSS-RM quasars found that, under relatively even and frequent sampling, the decisive factor was the line-to-continuum flux ratio rather than the sampling itself, with a critical ratio of about 6% above which $$f_{\rm BLR}(t)=\int d\tau\,\Psi(\tau)\,f_{\rm cont}(t-\tau),$$1 moved closer to unity and the scatter was markedly reduced (1711.02167). This is consistent with object-level studies in which H$$f_{\rm BLR}(t)=\int d\tau\,\Psi(\tau)\,f_{\rm cont}(t-\tau),$$2 fractions of 10–30% were described as usable but challenging, and H$$f_{\rm BLR}(t)=\int d\tau\,\Psi(\tau)\,f_{\rm cont}(t-\tau),$$3 contamination in continuum bands was treated as small but not identically zero (Ma et al., 2023).

Cadence and window functions generate a second cluster of systematics. Stripe 82 tests showed that seasonal gaps can drive lag solutions toward spurious values near $$f_{\rm BLR}(t)=\int d\tau\,\Psi(\tau)\,f_{\rm cont}(t-\tau),$$4 days when the radius–luminosity prior is removed (1711.02167). Microlensing-aided reverberation mapping simulations likewise found that seasonal gaps create spurious secondary peaks and bias lag estimates, and LSST-cadence experiments showed that two strategies with similar mean spacing and similar numbers of points can produce substantially different recoverability because of how they sample peaks and valleys in the intrinsic variability (Sluse et al., 2014, Jankov et al., 2021).

The stochastic process assumed for the continuum is another debated point. JAVELIN and much of the PRM literature adopt a DRW description, but its universality has been questioned. A high-redshift narrow-band PRM study of SDSSJ144645.44+625304.0 at $$f_{\rm BLR}(t)=\int d\tau\,\Psi(\tau)\,f_{\rm cont}(t-\tau),$$5 used many thousands of simulated CARMA process light curves and concluded that the input lag could still be recovered to within 6 per cent on average at the observed signal-to-noise of $$f_{\rm BLR}(t)=\int d\tau\,\Psi(\tau)\,f_{\rm cont}(t-\tau),$$6 and an average cadence of 14 days, even when DRW is not applicable (Read et al., 2019). By contrast, a modified broadband JAVELIN analysis on SDSS-RM and Stripe 82 data found that, for both SDSS-RM-like and Stripe 82-like light curves, the DRW model performed better than a power-law structure-function model in recovering broadband PRM lags (1711.02167). The practical implication is not that DRW is universally correct, but that its adequacy is dataset- and objective-dependent.

A recurrent misconception is that PRM eliminates spectroscopy altogether. It does not. Broad-band PRM generally does not resolve line profiles, so virial mass estimates still require an independent spectroscopic velocity width or a spectral estimate of the line fraction. The Wise NGC 4395 study explicitly used an adopted H$$f_{\rm BLR}(t)=\int d\tau\,\Psi(\tau)\,f_{\rm cont}(t-\tau),$$7 FWHM from spectroscopy, and the later minute-cadence NGC 4395 work used a single-epoch Gemini/GMOS spectrum to decompose the $$f_{\rm BLR}(t)=\int d\tau\,\Psi(\tau)\,f_{\rm cont}(t-\tau),$$8-band and measure broad H$$f_{\rm BLR}(t)=\int d\tau\,\Psi(\tau)\,f_{\rm cont}(t-\tau),$$9 width (Edri et al., 2012, Gu et al., 2024). Another misconception is that PRM is synonymous with broad-band BLR work; the literature includes narrow-band and medium-band BLR campaigns, continuum-only delay measurements, single-band lensed-quasar variants, and PRM-inspired polarimetric programs (Haas et al., 2011, Nuñez et al., 2019, Sluse et al., 2014, Shablovinskaya et al., 2023).

6. Survey-scale applications and methodological extensions

The large-survey motivation has been present since the earliest broad-band quasar formulations, but later work quantified its consequences. LSST-oriented PhotoRM simulations on 19 artificial light-curve pairs found that cadence design is a primary control variable: ideal cadence performed best, variable cadence slightly worse, and the AGN-friendly OpSim “agnddf” strategy outperformed the generic “baseline” strategy, with approximate MSE values of $R_{\rm BLR} \equiv c\tau,$00, $R_{\rm BLR} \equiv c\tau,$01, $R_{\rm BLR} \equiv c\tau,$02, and $R_{\rm BLR} \equiv c\tau,$03, respectively (Jankov et al., 2021). Continuum PRM simulations for LSST likewise concluded that delays can be recovered with an accuracy of 5 and 15% for light curves with a time sampling of 2 and 5 days, respectively, while emphasizing strong redshift dependence and sensitivity to line contamination and diffuse continuum emission (Nuñez et al., 2022).

The method has also been pushed toward high-redshift UV applications. A proposal to calibrate the C IV continuum size–luminosity relation with PRM argued that continuum lags can be obtained up to $R_{\rm BLR} \equiv c\tau,$04 faster than BLR sizes and that a 2-day cadence over a 6-month baseline should recover continuum delays with 10–15% accuracy for bright $R_{\rm BLR} \equiv c\tau,$05–3 quasars observed with the ESO La Silla 2.2 m telescope (Panda et al., 2024). In a different direction, microlensing-aided reverberation mapping showed that single-band photometry of lensed quasars can separate continuum and BLR variability because microlensing changes the continuum-to-BLR contrast between images; under fiducial daily sampling over 9 years, the recovered lag distribution remained centered near the true 100-day lag, and with $R_{\rm BLR} \equiv c\tau,$06 the uncertainty could be smaller than 10 days (Sluse et al., 2014).

Long-term small-telescope programs have demonstrated that PRM is not restricted to isolated proof-of-concept objects. A nearly decade-long H$R_{\rm BLR} \equiv c\tau,$07 PRM survey with 15–40 cm telescopes reported robust delays for 33/80 objects and added 31 objects to the H$R_{\rm BLR} \equiv c\tau,$08 size–luminosity relation, with a scatter of 0.26 dex within the new sample and 0.17 dex when the 6 lowest-luminosity sources were discarded (Figaredo et al., 2024). This suggests that, at least for bright nearby Seyferts with favorable lines, PRM has moved from a feasibility demonstration to a population-scale BLR characterization tool.

PRM has also become a template for adjacent methods. Medium-band polarimetric reverberation mapping explicitly adopted the PRM idea of a filter on the broad line and a nearby continuum filter, but applied it to polarized broad-line flux in order to probe equatorial scattering regions rather than the BLR itself (Shablovinskaya et al., 2023). A plausible implication is that PRM now functions as a general observational paradigm: use photometric contrast engineering, rather than full spectral resolution, to isolate a delayed component and turn large time-domain imaging datasets into size measurements of unresolved AGN substructures.