2412.06885v1

Draft version December 11, 2024
Typeset using LATEX twocolumn style in AASTeX631
The Sloan Digital Sky Survey Reverberation Mapping Project: Insights on Maximizing Efficiency in
Lag Measurements and Black-Hole Masses
Y. Homayouni,1, 2 Yuanzhe Jiang,3 W. N. Brandt,1, 2, 4 C. J. Grier,5 Jonathan R. Trump,6 Yue Shen,7, 8
Keith Horne,9 Patrick B. Hall,10 Scott F. Anderson,11 Luis C. Ho,12, 3 and D. P. Schneider13, 2
1 Department of Astronomy and Astrophysics, The Pennsylvania State University, 525 Davey Laboratory, University Park, PA 16802
2 Institute for Gravitation and the Cosmos, The Pennsylvania State University, University Park, PA 16802
3 Department of Astronomy, School of Physics, Peking University, Beijing 100871, China
arXiv:2412.06885v1 [astro-ph.GA] 9 Dec 2024
4 Department of Physics, 104 Davey Lab, The Pennsylvania State University, University Park, PA 16802, USA
5 Department of Astronomy, University of Wisconsin-Madison, Madison, WI 53706, USA
6 University of Connecticut, Department of Physics, 196 Auditorium Road, Unit 3046, Storrs, CT 06269-3046
7 Department of Astronomy, University of Illinois at Urbana-Champaign, Urbana, IL, 61801, USA
8 National Center for Supercomputing Applications, University of Illinois at Urbana-Champaign, Urbana, IL, 61801, USA
9 SUPA School of Physics and Astronomy, North Haugh, St. Andrews, KY16 9SS, Scotland, UK
10 Department of Physics and Astronomy, York University, Toronto, ON M3J 1P3, Canada
11 Astronomy Department, University of Washington, Box 351580, Seattle, WA 98195, USA
12 Kavli Institute for Astronomy and Astrophysics, Peking University, Beijing 100871, China
13 Dept. of Astronomy and Astrophysics, The Pennsylvania State University, 525 Davey Laboratory, University Park, PA 16802
ABSTRACT
Multi-year observations from the Sloan Digital Sky Survey Reverberation Mapping (SDSS-RM)
project have significantly increased the number of quasars with reliable reverberation-mapping lag
measurements. We statistically analyze target properties, light-curve characteristics, and survey design
choices to identify factors crucial for successful and efficient RM surveys. Analyzing 172 high-confidence
(“gold”) lag measurements from SDSS-RM for the Hβ, Mg II, and C IV emission lines, we find that
the Durbin-Watson statistic (a statistical test for residual correlation) is the most significant predictor
of light curves suitable for lag detection. Variability signal-to-noise ratio and emission-line placement
on the detector also correlate with successful lag measurements. We further investigate the impact of
observing cadence on survey design by analyzing the effect of reducing observations in the first year
of SDSS-RM. Our results demonstrate that a modest reduction in observing cadence to ∼1.5 weeks
between observations can retain approximately 90% of the lag measurements compared to twice-weekly
observations in the initial year. Provided similar and uniform sampling in subsequent years, this
adjustment has a minimal effect on the overall recovery of lags across all emission lines. These results
provide valuable inputs for optimizing future RM surveys.
Keywords: Active Galaxies, Quasars, Reverberation Mapping
1. INTRODUCTION a few SMBHs (GRAVITY Collaboration et al. 2018,

Mounting observations indicate a strong connec- 2020, 2021, 2024), spatially resolving the broad-line re-
tion between the formation and evolution of galaxies gion (BLR) of active galactic nuclei (AGN) remains a
throughout the Universe and the properties of super- challenge. Reverberation mapping (RM; Blandford &
massive black holes (SMBHs) residing at their centers. McKee 1982; Peterson 1993) allows measurement of cen-
Accurately determining black-hole masses (MBH ) and tral black-hole mass and study of the gas dynamics near
characterizing the nature of gas flows in their vicinity the black hole by temporally resolving the compact cen-
are crucial for understanding black-hole – galaxy coevo- tral regions. In its most general form, RM measures
lution and how black-hole-driven outflows can regulate the time delay τ (also known as the lag) between the
star formation through feedback mechanisms (e.g., Ko- continuum and emission-line variability signals in AGN
rmendy & Ho 2013). However, with the exception of to determine the responsivity-weighted radius of the re-
processing BLR region (RBLR ). This time delay can
2
be combined with the velocity dispersion of the emis- Time (LSST; e.g., Ivezić et al. 2019) produce millions of
sion line to estimate the black-hole mass using the virial light curves, developing strategic analysis tools will be
theorem (e.g., see the recent review by Cackett et al. crucial for timely and effective RM studies. Although
2021). this study focuses on emission-line RM, the methodol-
Three decades of RM campaigns have compiled a ogy developed here can be readily extended to investi-
set of ∼ 100 reliable RM masses in the local universe gate continuum RM, provided that a sufficient number
(e.g., Clavel et al. 1991; Wanders et al. 1997; Kaspi of continuum RM lag measurements are available.
et al. 2000; Collier et al. 1998; Peterson et al. 2004; In large-scale surveys like the SDSS-RM, a compre-
Bentz et al. 2009; Denney et al. 2010; Barth et al. hensive understanding of the interplay between the basic
2011; Grier et al. 2012; Du et al. 2014, 2016a,b; Barth quasar properties, the quasar intrinsic variability char-
et al. 2015; Hu et al. 2015). Most of these cam- acteristics, the survey design, and the observational sen-
paigns used single-object spectrographs, targeting the sitivity is critical for evaluating and forecasting the pro-
most variable, local AGNs (z <0.3), with relatively low- gram’s success in terms of lag-measurement yield and
luminosity (λL5100 < 1045 erg s−1 ) that would have suf- limitations. The detection of reverberation lags relies
ficiently short lags to be recovered with a few months heavily upon the monitoring program’s design, specif-
of RM monitoring. These campaigns also are predom- ically the observing cadence, total observing baseline,
inantly focused on strong emission lines such as the presence and distribution of seasonal/weather gaps, and
Balmer lines (see Bentz & Katz 2015 for a compilation the signal-to-noise ratio (S/N) of the flux measurements.
of MBH ). However, some campaigns have focused on Previous light-curve simulations of the SDSS-RM survey
high-z, luminous quasars using multiple years of RM have already studied the correlation between time-series
observations (Kaspi et al. 2007; Lira et al. 2018). Some analysis methods and survey yields (Li et al. 2019). The
of these campaigns also have been successful in perform- OzDES campaign has similarly employed light-curve
ing velocity-resolved RM observations (Bentz et al. 2008, simulations replicating source variability, measurement
2009, 2010; Grier et al. 2013; Bentz et al. 2021; Bao et al. errors, and observing cadence to assess lag measurement
2022; U et al. 2022; Zastrocky et al. 2024), and also us- reliability criteria (Penton et al. 2022). However, there
ing space-based observations in the UV (De Rosa et al. has never been a comprehensive assessment connecting
2015; Horne et al. 2021; Homayouni et al. 2023). physical sample properties and successful lag measure-
Recently, industrial-scale RM programs such as ment. The present work focuses on a sub-sample of
the Sloan Digital Sky Survey Reverberation Mapping the most reliable RM lag measurements from the SDSS-
(SDSS-RM) project (Shen et al. 2015a) and the Aus- RM survey, specifically those quasars exhibiting well-
tralian Dark Energy Survey RM (OzDES) project (King detected lag measurements (i.e., the “gold sample”, see
et al. 2015) have started to use multi-object spectro- § 3 for more details). By investigating the physical and
graphs on survey telescopes to carry out RM observa- statistical properties of this sub-sample, we aim to iden-
tions of large samples of diverse quasars. While these tify the characteristics of the already observed quasars
surveys have been successful, a significant fraction of that are most favorable for efficient RM lag measure-
the sample lacks well-detected lags. This is because ments from the SDSS-RM survey. We also explore the
extracting reliable and well-defined lag measurements implications of a reduced cadence in the SDSS-RM sur-
from industrial-scale RM surveys is challenging (see § 3 vey, enabling more efficient target selection, cadence de-
for more details). This is further complicated because, sign, and time-series analysis for future RM campaigns
at higher redshifts, observations are limited to more lu- like the 4MOST survey’s TiDES program (de Jong et al.
minous quasars, which tend to have longer time delays 2019; Swann et al. 2019), which aims to conduct RM ob-
due to the combined effects of cosmological time dilation servations of 700 quasars over five years, continuing to
and the canonical R − L relation (Bentz et al. 2013). monitor some of the previously studied RM fields.
Additionally, the inherent lower variability amplitude In § 2 we give an overview of the SDSS-RM sur-
observed in more luminous quasars (e.g.,Vanden Berk vey, sample selection, data, and data processing. § 3
et al. 2004; MacLeod et al. 2010) further reduces the describes our custom time-series analysis pipeline and
probability of successful lag recovery. Industrial-scale strategies for reliable lag identification, and the sample
RM faces challenges beyond observational costs due to used for this work. We present the analysis of physical
the massive volume of light curves. Efficiently analyzing AGN properties in § 4, and discuss connections to sta-
these light curves and selecting the most significant and tistical light-curve properties in § 5. In § 6 we describe
reliable lag measurements is computationally intensive. the impact of reduced cadence on lag success and re-
As future surveys like the Legacy Survey of Space and covery. We discuss the implications of our assessments
3
in § 7. Throughout this work, we adopt a ΛCDM cos- muchi et al. (2020) describe the photometric component
mology with ΩΛ = 0.7, ΩM = 0.3, and H0 = 70 km of the SDSS-RM program, and the associated data re-
s−1 Mpc−1 . duction. Furthermore, the SDSS-RM field overlaps with
the PanSTARRS-1 MD07 Medium Deep Field (Kaiser
2. DATA et al. 2010), providing additional multi-band photome-
2.1. SDSS-RM Survey Overview try (2010 – 2013). We also incorporate Zwicky Transient
Facility data (2018 – 2020, Bellm et al. 2019), which to-
The SDSS-RM project is a time-domain multi-object
gether with Pan-STARRS data, extend the light curve
spectroscopic RM (MOS-RM) survey that simultane-
baseline to 11 years (2010 – 2020) for investigation of
ously monitored 849 broad-line quasars at 0.1 < z < 4.5
longer lags in SDSS-RM observations.
in a single 7 deg2 field (Shen et al. 2015a). Due to
The spectroscopic data are initially processed using
its simple magnitude-limited selection criteria of ipsf ≤
the standard SDSS pipeline, followed by a custom cali-
21.7, the SDSS-RM survey has significantly expanded
bration pipeline to improve flux calibration (Shen et al.
the parameter space of AGNs studied with RM, which
2015a). The data are then further reprocessed using
allows for the investigation of luminous quasars beyond
the PrepSpec software (Shen et al. 2015a, 2016) to im-
the local universe. An overview of the SDSS-RM de-
prove relative flux calibration assuming the flux of nar-
sign, observing strategy, and target selection is reported
row emission lines do not intrinsically vary throughout
in Shen et al. (2015a), with details of sample properties
the RM campaign. PrepSpec then models the spectra,
reported in Shen et al. (2019a). The primary goal of
continuum, and the broad line and produces light curves.
SDSS-RM is to provide SMBH mass measurements for
PrepSpec also produces measurements of mean and root
a broad range of redshifts and luminosities (Shen et al.
mean square (rms) residual line profiles, line widths, and
2016; Grier et al. 2017, 2019; Homayouni et al. 2020;
light curves for each of the model components. It also
Shen et al. 2024). However, it also has been successful
returns several other statistical products (see Section 5
in enabling many ancillary studies of quasar variability
for more details).
and other physical properties (e.g., Shen et al. 2015b;
The photometric data from different instruments, fa-
Sun et al. 2015; Denney et al. 2016a,b; Grier et al. 2016;
cilities, and filters are combined to account for ob-
Dexter et al. 2019; Hemler et al. 2019; Homayouni et al.
servatory seeing variations, calibration issues for each
2019; Wang et al. 2019; Li et al. 2019; Fonseca Alvarez
filter response, telescope throughput, and other site-
et al. 2019; Dalla Bontà et al. 2020; Li et al. 2021, 2023;
dependent effects. We adopt the publicly available
Fries et al. 2023).
PyCali code (Li et al. 2014) to perform the light curve
2.2. Data Processing merging process. PyCali uses a Bayesian framework
to achieve this, allowing for the adjustment of individ-
SDSS-RM observed every year during 2014–2020 as
ual light curve flux uncertainties. This process miti-
part of SDSS-III (Eisenstein et al. 2011) and SDSS-IV
gates overestimation and underestimation of the uncer-
(Blanton et al. 2017), and observations of a subset of
tainties reported in the light curves. We normalize all
the SDSS-RM quasar field are continuing as part of the
light curves to the flux of synthetic photometry in the
SDSS-V Black Hole Mapper (BHM) program. The spec-
r-band, where we measure the synthetic photometry by
troscopic monitoring required for SDSS-RM was pro-
convolving the PrepSpec-corrected spectra with SDSS
vided by the BOSS spectrograph (Smee et al. 2013) on
filters responses (Fukugita et al. 1996) to establish a
the SDSS telescope (Gunn et al. 2006). The observations
common reference flux level.
achieved an average cadence of 4 days with 32 epochs
in the first year, ∼ 12 epochs per year between 2015
– 2017, and ≈ 6 epochs per year (monthly cadence) 2.3. Sample Selection
during 2018 – 2020, totalling 90 spectroscopic epochs In less than one decade, the SDSS-RM program has
over the course of seven years of monitoring. SDSS-RM substantially expanded the set of reliable SMBH masses
was also accompanied by optical photometric monitor- measured through RM to ∼300 quasars out to z >3
ing in the g and i-band from the 2.3 m Bok telescope (Shen et al. 2016; Grier et al. 2017, 2019; Shen et al.
at Steward Observatory and the 3.6 m Canada France 2019b; Homayouni et al. 2020; Shen et al. 2024). Grier
Hawaii Telescope (CFHT) to enhance the continuum et al. (2017) measured Hβ and Hα lags from the first-
light curve to facilitate RM measurements. These obser- season observations from 2014. For longer lags in Hα,
vations largely overlap with the spectroscopic observa- Hβ, Mg II, and C IV, multi-year RM lag results were
tion window, and have similarly higher cadence in 2014, reported by Grier et al. (2019), Shen et al. (2019b),
and reduced cadence in the subsequent years. Kine- Homayouni et al. (2020), and Shen et al. (2024). In this
4
work, we focus on the targets with the highest-quality model-dependent methods, particularly in industrial-
lag measurements from these previous SDSS-RM stud- scale datasets like the SDSS-RM program and similar
ies. Our subsample is drawn from those SDSS-RM stud- reverberation-mapping campaigns (Li et al. 2019; Yu
ies that use the improved continuum light curves from et al. 2020). While established methods like JAVELIN
ground-based photometry. We select lag measurements and ICCF have been well-studied for reverberation map-
that have quality ratings of 4 or 5 from Grier et al. ping, more recent techniques like PyROA lack comprehen-
(2017, 2019); Shen et al. (2024) (also known as the gold sive comparisons with these existing methods.
sample) and the most reliable lag measurements with
individual false-positive rate (FPR) of < 10% as defined 3.1. Alias Removal
in Homayouni et al. (2020). In total, we have 137 RM
Sparsely-sampled data in RM time-series analysis can
lag measurements that are flagged as the “gold sam-
lead to artifacts in lag-measurement results. Specifi-
ple” with 26 lag measurements using Hβ (Grier et al.
cally, posterior lag distributions obtained from meth-
2017), 24 from Mg II (Homayouni et al. 2020), and 16
ods like JAVELIN and CREAM, and the cross-correlation
in C IV (Grier et al. 2019) reported from the early-year
centroid distribution (CCCD) from ICCF, may gener-
SDSS-RM studies, and 37 in Hβ, 32 in Mg II, and 37 in
ate a pronounced primary peak that is accompanied
C IV from a recent SDSS-RM investigation (Shen et al.
by the presence of secondary peaks, generally less sig-
2024). In this work, the one-year results from Grier et al.
nificant than the primary. These secondary peaks are
(2017), and four-year results from Grier et al. (2019) and
likely aliases, arising from the interplay of low-cadence
Homayouni et al. (2020) are referred to as the “early-
sampling and noisy measurements, where the Markov
year” SDSS-RM lag results. We refer to the most recent
chain Monte Carlo (MCMC) algorithm, in attempting
lag measurements from Shen et al. (2024) combining 7
to fit the light curve, may introduce spurious correla-
years of spectroscopy with 11 years of photometry as
tions in weakly variable segments, leading to these sec-
the “7-year” SDSS-RM results. We also refer to the
ondary peaks. The presence of multiple spurious peaks,
larger SDSS-RM sample of 849 quasars as the “parent
or aliases, within the lag posterior distribution func-
sample”.
tion (PDF) can significantly compromise lag estimation.
While one alias may appear dominant, the presence of
3. TIME-SERIES ANALYSIS others indicates a poorly constrained lag measurement.
Ideally, RM observations rely on densely-sampled data This skews the PDF, potentially leading to biased lag es-
to precisely track AGN variability and perform ro- timates and inflated uncertainties. The presence of sea-
bust time-series analysis. However, observing limita- sonal gaps further complicates aliasing in RM analysis.
tions such as weather loss and telescope-scheduling con- Methods like JAVELIN model the light curve throughout
straints can lead to sparsely sampled light curves. This the entire series, including seasonal gaps, which can lead
sparsity in a light curve requires interpolation between to spurious lag solutions that coincide with the gaps.
observing epochs to accurately measure lags and their This occurs because the algorithms may misinterpret
associated uncertainties. Variability tracking is further the absence of data during seasonal gaps as a correlated
challenged in multi-year survey observations because of signal, leading to artificial peaks in the lag posterior dis-
the presence of seasonal gaps in the data. Historically, tribution. Therefore, to mitigate the effects of aliasing
RM observations use the Interpolated Cross Correlation in the lag PDF, a systematic approach to removing these
Function (ICCF) method to perform time-series analy- artifacts is necessary.
sis (Gaskell & Sparke 1986; Gaskell & Peterson 1987; Prior studies within the SDSS-RM collaboration have
Peterson et al. 2004). To perform time-series anal- employed a consistent alias removal approach. This
ysis for SDSS-RM light curves, we adopt approaches method incorporates a weighting function on the lag
that are flexible in handling multi-year observations PDF that assigns lower weights to lag values correspond-
that are associated with observing gaps: JAVELIN (Zu ing to segments with minimal overlap between the con-
et al. 2011), CREAM (Starkey et al. 2016), and more re- tinuum and line light curves (Grier et al. 2017, 2019;
cently PyROA (Donnan et al. 2021). These methods Homayouni et al. 2020; Shen et al. 2024). Essentially,
model the light curve behavior during observing gaps lags exhibiting a complete absence of overlap in the light
assuming stochastic variability of quasar light curves, curves at a specific time delay (τ ) are deemed less prob-
whereas the ICCF method relies on linear interpolation able by the weighting scheme. The weighting scheme
between epochs. Comparison between these methods incorporates two key factors to address aliasing: the
reveals that the ICCF technique results in higher un- number of overlapping data points and the influence of
certainty in lag measurements compared to the other continuum variability; the final weighting scheme is the
5
convolution between these two terms. The first factor is • Lag |S/N|: We can assess the consistency of a lag
captured by [N (τ )/N (0)]2 , where N (τ ) represents the measurement with zero lag by considering the ab-
number of overlapping data points at a specific time solute value of the measured lag and its associated
lag and N (0) denotes the number of overlapping points uncertainty, and can identify if a lag measurement
at zero lag τ = 0. Lower values of this term indicate is consistent with zero.
minimal overlap. The second factor is introduced by
the continuum auto-correlation function (ACF). A nar-
row ACF signifies a rapidly varying continuum, making • rms variability S/N: This criterion effectively re-
it challenging to distinguish the light curve’s behavior moves cases where the light curves exhibit min-
within the gaps. Conversely, a broad ACF indicates a imal intrinsic variability, thereby preventing the
slowly varying continuum, where gaps have a minimal lag detection methods from erroneously identify-
impact on the measured lag. ing monotonic trends or spurious correlations be-
tween noisy light curves.
3.2. Significance Criteria
While the alias-removal approach eliminates spuri-
ous peaks and aliases from the lag PDF, further as- These are our general selection criteria, though
sessment is necessary to identify the most reliable lag some studies have adopted more stringent requirements
measurements. This becomes important since the alias (Homayouni et al. 2019).
removal process might inadvertently suppress genuine
peaks alongside aliases, potentially leaving the alias-
rejected lag PDF with a weak primary peak that is dif- 3.3. Selection of the “Gold Sample”
ficult to confidently identify. Furthermore, the lag PDF Large-scale RM surveys like the SDSS-RM program
may identify lags that are statistically consistent with are inherently susceptible to a certain level of false-
zero, offering no meaningful physical interpretation. To positive lag detections. Limited sampling cadence and
address these concerns, we employ additional selection the presence of seasonal gaps can lead to lag PDFs ex-
criteria to ensure the final reported lags are statistically hibiting well-defined peaks that satisfy our significance
robust and correspond to a true physical reverberation criteria. These peaks may arise from noisy light curves
process. or spurious correlations within the light curve, rather
To identify statistically significant lag measurements, than reflecting a true physical reverberation process.
we employ a set of selection criteria established through Therefore, Grier et al. (2017); Homayouni et al. (2020);
a combined approach of statistical assessment and visual Grier et al. (2019) and Shen et al. (2024) have employed
inspection of the lag PDF and the light curves. Ideally, a secondary selection process to identify the most reli-
these criteria are chosen to achieve a false-positive rate able lag measurements, often referred to as the gold sam-
of ≈ 10%. This target rate is informed by simulations ple. This selection process has primarily relied on visual
from Shen et al. (2015a) and considers the data quality inspection of the light curves and lag PDFs to identify
within the SDSS-RM program, including factors such these high-confidence lag measurements. Distinguishing
as light-curve cadence, signal-to-noise ratio (S/N), and genuine reverberation from spurious detections remains
the presence of seasonal gaps. Previous SDSS-RM lag a challenge, with the specific criteria varying between
measurements have employed a suite of criteria to assess studies. Key factors for gold-lag measurement include a
the significance of lag detections, which include: unimodal lag probability density function, indicating a
single dominant lag value, and sufficient light curve vari-
• fpeak : The fraction of the weighted lag posterior
ability. Furthermore, consistency across different analy-
integrated within the primary peak can be used
sis methods and a good fit of the applied model to the
as a criterion for significance. This metric assesses
light curve data provide strong supporting evidence for
whether the primary peak concentrates a sufficient
the validity of the measured lag. Other approaches such
fraction of the posterior probability to be consid-
as simulation-based methods and quantitative measures
ered a reliable lag measurement.
can provide objective assessments of lag reliability, but
• rmax : The maximum Pearson cross-correlation co- they often require significant computational resources,
efficient rmax between the continuum and the line as demonstrated by Homayouni et al. (2020) and Yu
light curves can serve as an indicator of correlated et al. (2021). This is especially true for identifying a
variability and physical time lag due to reverber- gold sample in industrial RM studies, which involves
ation processes. time-intensive simulations of lightcurves.
6
Early years SDSS-RM 7-year SDSS-RM Comparison
47
Median log λLλ 5100 (erg s−1)
Median log λLλ 5100 (erg s−1)

SDSS-RM SDSS-RM
48 Grier+17 (Hβ) 48 Shen+24 (Hβ)
log λLλ 5100 (erg s−1)
Homayouni+20 (MgII) Shen+24 (MgII) 46

Grier+19 (CIV) Shen+24 (CIV)
46 46 45
44 44 44
0 2 4 0 2 4 44 46
z z log λLλ 5100 (erg s−1)
1.00 1.00
0.75 0.75
Success Ratio
Success Ratio
0.50 0.50
0.25 0.25
0.00 0.00
44 46 42 44 46
log Lλ 5100 log Lλ 5100
Figure 1. Comparison of the host subtracted λL5100 continuum luminosity in the gold sample of SDSS-RM quasars. The
SDSS-RM parent sample is illustrated in gray, open symbols. The top left panel illustrates the gold lag measurements (colored
symbols) from the first-year and four-year (early results) SDSS-RM measurements in Grier et al. (2017); Homayouni et al.
(2020); Grier et al. (2019). The top middle panel shows the 7-year lag measurements in Hβ, Mg II, and C IV from Shen et al.
(2024). The continuum luminosity here are host-subtracted using the principle component analysis as reported by Shen et al.
(2015b). Overall, there is no notable difference between the luminosity in the SDSS-RM parent sample and the targets with
the most reliable lag measurements in either the early SDSS-RM or 7-year lag measurements (see Table 2 for a comparison of
median values in each work). The top right panel compares the first-year with the 7-year continuum luminosity in the SDSS-RM
sample. Overall, the targets do not show a significant change in their luminosity. The bottom panels show the gold-lag success
fraction as a function of the early-year (left) and 7-year (right) continuum luminosity. The lag success ratio exhibits a decreasing
trend with increasing source luminosity. This could be due to the longer time delays of more luminous targets (e.g., Bentz et al.
2013), which may require longer observational baselines not covered by the studies in the current work.
Table 1. Gold Lag Summary in SDSS-RM Survey
Targeted Emission Line Results NSample Total NGold Gold-Lag Success Ratio
Hβ (first-year) Grier et al. (2017) 222 26 12%

MgII (four-year) Homayouni et al. (2020) 193 24 12%
CIV (four-year) Grier et al. (2019) 349 16 5%
Hβ (seven-year) Shen et al. (2024) 186 37 20%
MgII (seven-year) Shen et al. (2024) 714 32 4%
CIV (seven-year) Shen et al. (2024) 494 37 7%
7
200 200 200
Hβ SDSS-RM (First Year) MgII SDSS-RM (4-year) CIV SDSS-RM (4-year)
Grier+17 (Hβ) Homayouni+20 (MgII) Grier+19 (CIV)
150 150 150
MgII REW (Å)

Hβ REW (Å)
CIV (Å)
100 100 100
50 50 50
00.0 0.5 1.0 0 0.5 1.0 1.5 0 2 3 4

z z z
0.5 0.5 0.5
Success Ratio
Success Ratio
Success Ratio
0.0 0.0 0.0

0 50 100 150 200 0 50 100 150 200 0 50 100 150 200
Hβ REW (Å) MgII REW (Å) CIV REW (Å)
200 200 200

Hβ SDSS-RM (7-year) MgII SDSS-RM (7-year) CIV SDSS-RM (7-year)
Shen+24 (Hβ) Shen+24 (MgII) Shen+24 (CIV)
150 150 150

MgII REW (Å)
CIV REW (Å)

Hβ REW (Å)
100 100 100
50 50 50
00.0 0.5 1.0 0 1 2 0 2 3 4

z z z
Success Ratio
Success Ratio
Success Ratio
0.4 Shen+24 (Hβ) 0.4 Shen+24 (MgII) 0.4 Shen+24 (CIV)
0.2 0.2 0.2
0.0 0.0 0.0

0 50 100 150 200 0 50 100 150 200 0 50 100 150 200
Hβ REW (Å) MgII REW (Å) CIV REW (Å)
Comparison Comparison Comparison
200 200 200
Hβ SDSS-RM
MgII SDSS-RM
CIV SDSS-RM
7-year MgII REW (Å)

7-year CIV REW (Å)
7-year Hβ REW (Å)
150 150 150
100 100 100
50 50 50
00 100 200 00 100 200 00 100 200

First-year Hβ REW (Å) 4-year MgII REW (Å) 4-year CIV REW (Å)
Figure 2. Comparison of the SDSS-RM REW for Hβ (left column), Mg II (middle column), and C IV (right column). The gold
sample in each study is identified by the colored stars and the parent sample in each case is shown with grey symbols, which
reveals a uniform distribution when compared to the parent population. For each work, we also include a panel that illustrates
the gold-lag success fractions as a function of REW, which reveals that there is no discernible trend evident for the success ratio
and larger REW. The panels in the bottom row show the REW comparison between the early-year and the 7-year lag results.
Table 2 demonstrates that gold lags in the C IV sample have marginally larger REWs compared to the other emission lines.
While these larger REWs may be expected to influence the gold-lag success ratio, the observed correlation is relatively weak.
8
From the parent sample of 849 quasars in the SDSS- 2012; Vanden Berk et al. 2004), potentially reducing
RM sample, the applicable redshift range for Hβ lags the likelihood of detecting a reliable lag measurement
(0.3 < z < 1.14) has limited the usable sample to 222 ob- within short observing campaigns. To explore this con-
jects. From this subsample, Grier et al. (2017) achieved nection, we adopt the λL5100 luminosity as a measure
successful lag measurements for 44 quasars, with 26 clas- of the quasar continuum luminosity. Whenever possi-
sified as gold. Similarly, RM measurements of Mg II in ble, we utilize directly measured λL5100 values. For
Homayouni et al. (2020) were restricted to 0.3 < z < 1.7, sources lacking direct λL5100 measurements, we recon-
leading to 193 quasars in the Mg II-subsample, which struct them using λ L3000 , λ L1350 , and the bolometric
yielded successful lag measurements in 57 quasars, out corrections from Richards et al. (2006). Figure 1 illus-
of which 24 were identified as gold. Grier et al. (2019) trates the redshift and luminosity distribution of our tar-
investigated C IV lags, but their analysis was limited gets. We performed a global comparison of the contin-
to 1.4 < z < 4.5 due to the upper redshift limit of the uum luminosity within the gold samples of early SDSS-
SDSS-RM parent sample. Out of 349 quasars, success- RM studies and the current 7-year lag measurements
ful lag measurements were obtained for 48, only 16 of (Grier et al. 2017, 2019; Homayouni et al. 2020; Shen
which were classified as gold. The resource-intensive na- et al. 2024). This analysis did not reveal any statis-
ture of RM, coupled with the low yield of gold-lag mea- tically significant changes in the continuum luminosity
surements, motivates further investigation in the current distribution over the multi-year analysis. Additionally,
work. Similarly for the 7-year lag results, Shen et al. we find no dependence between lag measurement success
(2024) recently analyzed a sample of 187 quasars for and source luminosity. The median continuum luminos-
Hβ, 714 for Mg II, and 494 for C IV, identifying 37, 32, ity of the gold sample (in logarithmic scale) is 44.1 ± 0.1
and 37 gold measurements, respectively. Table 1 pro- erg s−1 , 44.7 ± 0.1 erg s−1 , and 45.0 ± 0.2 erg s−1 for Hβ,
vides a brief summary of the overall success fraction of Mg II, and C IV respectively for the early SDSS-RM
gold-lag measurements in SDSS-RM. lag measurements, which is similar to the median range
of 44.2 ± 0.2, 44.8 ± 0.05, and 45.0 ± 0.1 erg s−1 in the
4. TARGET PROPERTIES parent sample of each targeted emission line. Here the
We examine the extent to which inherent target prop- uncertaintly on the median vales are computed through
erties, including both physical characteristics and sta- bootstrap method. We find similar ranges for the gold-
tistical light-curve signatures, produce the most reliable lag measurements in the 7-year results of Shen et al.
lag measurements. To ensure that target properties do (2024).
not vary significantly relative to the baseline observa-
tions, we study the properties within each study. These 4.2. Rest-frame Equivalent Width
included one-year and four-year RM lag measurements The rest-frame equivalent width (REW) of emission
reported by Grier et al. (2017); Homayouni et al. (2020), lines is another potential factor influencing successful
and Grier et al. (2019) alongside the recent seven-year lag measurement success. We investigate whether tar-
results by Shen et al. (2024). This comparison aims to gets with larger REWs correlate with the most reliable
mitigate any significant discrepancies potentially aris- lag measurements. For each emission line, we examine
ing from the difference in the baseline of observations the potential relationship between its REW using Prep-
(short vs. long). For each property, we also investigate Spec outputs. REWs are measured from the continuum
the correlation between gold-lag success ratio and the and flux provided by PrepSpec, utilizing either one-year,
property’s value. A comparative analysis of the median four-year, or seven-year average spectra. Figure 2 shows
values for the parent and gold samples is also provided the comparisons between early-year and 7-year SDSS-
in Table 2 (physical target properties) and Table 3 (sta- RM data for the same emission lines, which reveal rel-
tistical light-curve properties). atively consistent REWs over time. Furthermore, we
find no direct correlation between the fraction of quasars
4.1. Continuum Luminosity with gold-lag measurements. Figure 2 depicts how the
Intrinsic AGN luminosity is considered the primary gold samples in each study are uniformly distributed
driver of reverberation lag times. More luminous within their respective parent samples, and the gold-lag
quasars tend to have larger time lags as observed by success fraction does not reveal any specific trends with
the R − L relation (Bentz et al. 2013; Fonseca Alvarez REW. While the one-year and four-year PrepSpec spec-
et al. 2019), which√matches basic photoionization expec- tral fits did not incorporate any Fe II emission lines,
tations (RBLR ∝ L). Furthermore, luminous quasars the seven-year data included an Fe II template. Con-
often exhibit lower amplitude variability (MacLeod et al. sequently, the 7-year measurements of Shen et al. 2024
9
Table 2. Comparative Analysis of Sample Properties: Gold and Parent
Emission Line λL5100 REW

log(erg s−1 ) Å
Parent Gold Parent Gold
Hβ (first-year) 44.37+0.37
−0.43 44.14+0.24
−0.42 44.83+66.95
−30.33 42.94+27.09
−20.19
MgII (four-year) 44.73+0.5
−0.41 44.6+0.2
−0.29 44.73+26.02
−14.9 52.19+16.27
−17.09
CIV (four-year) 45.09+0.4
−0.41 44.86+0.33
−0.32 48.0+43.49
−23.73 79.77+17.17
−38.09
Hβ (seven-year) 44.32+0.42
−0.43 44.15+0.31
−0.33 54.07+26.83
−24.9 54.36+23.4
−15.25
MgII (seven-year) 44.75+0.49
−0.42 44.65+0.24
−0.35 49.3+23.87
−17.59 55.83+27.31
−19.73
CIV (seven-year) 45.03+0.39
−0.42 44.93+0.37
−0.34 47.69+37.29
−20.04 67.47+28.65
−23.68
(blue data points in Figure 2) appear to be systemati- wavelength range of the detector. The results for C IV
cally offset above the 1:1 line, suggesting that for all the are less conclusive. The highest redshift in the C IV
three emission lines, the gold-lag estimates may be bi- gold sample (z = 2.9) restricts our ability to confirm
ased towards larger REWs. However, we did not observe the lag success rate beyond ≈6000 Å. Additionally, as-
a higher success rate for the gold-lag sample, except for sessing C IV results is complicated as these targets tend
C IV. to have longer rest-frame lags due to their higher lu-
minosity and cosmological time dilation effects. Our
4.3. Redshift and Observed-Frame Wavelength comparison across the three emission lines suggests a
clear influence only for Hβ where the three central bins
The SDSS-RM quasars probe 0.3 < z < 4.5; red-
covering 5725 Å to 7425 Å exhibit a mean success rate
shift effects limit the observable emission lines that
of ≳20%. The central bins from Shen et al. (2024) ex-
can be targeted with RM observations and cosmological
hibited a higher success fraction of approximately 39%
time dilation increases the observed lags. The through-
compared to the approximately 30% success rate of the
put of the SDSS BOSS spectrograph exhibits a wave-
central bins from Grier et al. (2017) and the neighbor-
length dependence across its operational range of 360
ing bins closer to the spectrograph edges (outer thirds
nm < λ < 1000 nm (Smee et al. 2013). This varia-
compared to the middle third). The results for Mg II
tion is attributed to several factors, including the CCD
and C IV were less definitive.
sensitivity, the blue/red grism efficiency, the collima-
tor characteristics, and contamination by sky lines. We
next explore the potential influence of the targeted emis- 5. LIGHT-CURVE VARIABILITY
sion line’s location within the SDSS spectrograph band- CHARACTERISTICS
pass on the gold-lag measurements. The detector edges
exhibit lower sensitivity, resulting in increased noise. 5.1. Fractional Variability
Therefore, we investigate whether the placement of the Quasars exhibit variability characterized by a wide
emission line closer to the detector center improves the range of amplitudes, wavelengths, and timescales (e.g.,
probability of acquiring a successful lag measurement. Collier et al. 2001; Peterson et al. 2004; Kelly et al.
Considering the redshift range of the SDSS-RM sam- 2009; MacLeod et al. 2012). This variability is crucial
ple (Section 2.1), prominent BLR emission lines observ- for RM studies. However, not all quasars exhibit suffi-
able include Hβ at lower redshifts, Mg II in the mid- cient variability of their emission lines to enable reliable
redshift range, and C IV at higher redshifts. We inves- RM measurements. It is plausible that the success of
tigate the impact of target redshift and observed-frame lag measurements is linked to specific characteristics of
wavelength for each emission line by dividing the band- the light-curve variability. Shen et al. (2024) quantified
pass into seven bins of equal observed wavelength range. the intrinsic variability of SDSS-RM light curves using
Figure 3 compares the success ratio of the observed- a maximum-likelihood estimator. They report the in-
frame emission line on the detector for both early-year trinsic root-mean-square (rms) variability for both the
SDSS-RM and the 7-year lag measurements. Our anal- 11-year photometric light curve and the 7-year emission-
ysis reveals a trend of decreasing lag success with wave- line light curves. Their findings indicate that different
length for Hβ, and Mg II. We also find a significantly emission lines exhibit varying degrees of intrinsic vari-
higher success rate for the gold Hβ lag measurements ability, although a general correlation exists between the
when the observed frame Hβ is located near the mid- continuum and emission-line rms variability.
10
0.6 0.6 0.6
0.5
Hβ 0.5
MgII 0.5
CIV
5 14
6 6
Success Ratio
0.4 0.4 0.4
0.3 10 0.3 0.3

3
0.2 7 0.2 0.2
5 19
8 9 9
0.1 4 0.1 3 3 6 7 5 0.1 146 4
2 1 3 2 1 8 2
0.04000 6000 8000 10000 0.04000 6000 8000 10000 0.04000 6000 8000 10000
Wavelength [Å] Wavelength [Å] Wavelength [Å]
Figure 3. Lag success fraction as a function of emission-line position on the BOSS spectrograph. Early SDSS-RM measurements
are depicted in reddish tones, while the 7-year lag measurements are represented by bluish tones. For Hβ (left panel), we have
the highest ratio of obtaining a successful lag measurement when the emission line is positioned near the center of the detector
(middle thirds), and we see a significant decrease in the lag recovery rate as the Hβ (left) profile falls on the edges (outer thirds)
of the detector both in the single-year (Grier et al. 2017) and the 7-year (Shen et al. 2024) analyses. The effect of emission-line
location on the detector in Mg II (middle) and C IV measurements (right) are more mixed; this is because the 4-year Mg II
study in Homayouni et al. (2020) has been cut at z <1.7 due to contamination with skylines. The C IV emission line can be
measured for targets at higher redshift and the lags are plausibly longer than could be recovered with the observed baseline of
4-years (Grier et al. 2019) or 7-years (Shen et al. 2024). For each wavelength bin, we also display the number of gold lags to
illustrate the statistical significance in each bin.
Table 3. Comparative Analysis of Light Curve Properties: Gold and Parent
Emission Line Frac. RMS Var. SNR2 Con. dw Line dw
Parent Gold Parent Gold Parent Gold Parent Gold

Hβ (first-year) 0.1+0.24
−0.06 0.1+0.14
−0.05 10.21+10.08
−4.79
+15.68
13.86−4.1 0.76+0.56
−0.5 0.3+0.5
−0.15 1.51+0.4
−0.61 0.83+0.6
−0.35
MgII (four-year) 0.1+0.06
−0.04 0.1+0.04
−0.03 29.48+11.67
−6.7
+11.71
25.88−3.48 0.38+0.52
−0.24 0.43+0.25
−0.33 1.17+0.36
−0.43 1.04+0.44
−0.5
CIV (four-year) 0.11+0.07
−0.05 0.14+0.09
−0.06 32.96+18.49
−9.46
+15.18
42.34−10.86 0.53+0.52
−0.31 0.4+0.48
−0.17 0.96+0.39
−0.39 0.54+0.32
−0.09
Hβ (seven-year) 0.16+0.25
−0.08 0.18+0.14
−0.09 23.98+28.77
−9.69 51.4+27.15
−19.9 0.62+0.43
−0.35 0.47+0.39
−0.25 1.0+0.52
−0.57 0.39+0.38
−0.19
MgII (seven-year) 0.1+0.07
−0.04 0.1+0.13
−0.04 22.33+14.88
−8.6 36.3+14.62
−11.57 0.67+0.41
−0.36 0.38+0.38
−0.2 1.39+0.31
−0.42 0.78+0.38
−0.37
CIV (seven-year) 0.12+0.08
−0.05 0.13+0.06
−0.07 33.09+21.39
−14.07 46.7+28.63
−13.8 0.67+0.42
−0.33 0.6+0.37
−0.27 1.05+0.38
−0.41 0.64+0.37
−0.21
We investigate the connection between emission-line et al. (2020) were cut at a redshift of z = 1.7 due to
rms variability and the success rate of gold-lag measure- the contamination by skylines, therefore, there are no
ments. We utilize the fractional rms variability (nor- lag measurements for Mg II at z > 1.7 based on the
malized to the mean flux) measured by PrepSpec as early SDSS-RM study (Homayouni et al. 2020) and the
described in Shen et al. (2019a). Our analysis reveals 7-year measurements reported by Shen et al. (2024) only
a broad distribution of fractional rms variability within includes three targets at z> 1.7.
the gold sample. Significant sky line residuals contami- The results of this investigation are presented in Fig-
nate the Hβ fractional RMS variability, potentially lead- ure 4. We further compared the fractional rms variabil-
ing to overestimated values (Shen et al. 2019a). While ity between the early-year and 7-year studies for each
the effect of skylines is less pronounced for Mg II, it is emission line bottom panels in Figure 4. It is notewor-
still present at z > 1.5 among the lag measurements thy that the 4-year and 7-year results largely follow a
of Shen et al. (2024), where skylines contaminate the 1:1 trend. The deviation observed in the comparison be-
spectra. The Mg II lag measurements in Homayouni tween the 1-year and 7-year Hβ fractional rms variability
11
1.0 1.0 1.0
SDSS-RM (First Year) SDSS-RM (4-year) SDSS-RM (4-year)
Hβ Grier+17 (Hβ) MgII Homayouni+20 (MgII) CIV Grier+19 (CIV)
MgII Fractional RMS Variability
CIV Fractional RMS Variability

Hβ Fractional RMS Variability 0.8 0.8 0.8
0.6 0.6 0.6
0.4 0.4 0.4
0.2 0.2 0.2
0.0
0.00 0.25 0.50 0.75 1.00 0.0 0.5 1.0 1.5 0.0 2 3 4
z z z
1.0 1.0 1.0
Success Ratio
Success Ratio
Success Ratio
0.5 0.5 0.5
0.0 0.0 0.0

0.0 0.2 0.4 0.6 0.8 1.0 0.0 0.2 0.4 0.6 0.8 1.0 0.0 0.2 0.4 0.6 0.8 1.0
Hβ Fractional RMS Variability MgII Fractional RMS Variability CIV Fractional RMS Variability
1.0 1.0 1.0

SDSS-RM (7-year) SDSS-RM (7-year) SDSS-RM (7-year)
Hβ Shen+24 (Hβ) MgII Shen+24 (MgII) CIV Shen+24 (CIV)
MgII Fractional RMS Variability
CIV Fractional RMS Variability

Hβ Fractional RMS Variability
0.8 0.8 0.8
0.6 0.6 0.6
0.4 0.4 0.4
0.2 0.2 0.2
0.0
0.00 0.25 0.50 0.75 1.00 0.0 0.5 1.0 1.5 2.0 2.5 0.0 2 3 4
z z z
1.0 1.0 1.0
Success Ratio
Success Ratio
Success Ratio
0.5 0.5 0.5
0.0 0.0 0.0

0.0 0.2 0.4 0.6 0.8 1.0 0.0 0.2 0.4 0.6 0.8 1.0 0.0 0.2 0.4 0.6 0.8 1.0
Hβ Fractional RMS Variability MgII Fractional RMS Variability CIV Fractional RMS Variability

1.0 1.0 1.0
SDSS-RM SDSS-RM SDSS-RM
Hβ MgII CIV
7-year MgII Fractional RMS Variability
7-year CIV Fractional RMS Variability

7-year Hβ Fractional RMS Variability

0.8 0.8 0.8
0.6 0.6 0.6
0.4 0.4 0.4
0.2 0.2 0.2
0.0 0.00.0 0.00.0

0.00 0.25 0.50 0.75 1.00 0.2 0.4 0.6 0.8 1.0 0.2 0.4 0.6 0.8 1.0
First-year Hβ Fractional RMS Variability 4-year MgII Fractional RMS Variability 4-year CIV Fractional RMS Variability
Figure 4. Fractional root-mean-square (rms) variability for Hβ (left column), Mg II (middle column), and C IV (right column)
as a function of redshift for early (reddish color palette) and 7-year (blueish tones) SDSS-RM results. For each work, we also
include a panel that illustrate the gold-lag success fraction as a function of fractional RMS variability with similar color palettes,
which reveals no trends between the gold-lag success ratio and fractional rms variability for the corresponding emission line (see
Table 3 for a comparison of median values in each work). The bottom panels compare the fractional rms variability between the
early-year and 7-year SDSS-RM campaigns. Our analysis reveals no statistically significant correlation between the emission
line fractional rms variability and gold-lag measurements.
12
(bottom left corner of Figure 4) can be attributed to the for all lag measurements in their dataset, regardless of
variability of quasars on the observed-frame timescales. whether they were classified as gold (see Figure 16 of
The damping timescale of quasars are longer than the Shen et al. (2024)). Similar to fractional RMS variabil-
seasonal monitoring duration. ity, SNR2 inherently increases with longer light-curve
baselines. We therefore refrain from showing compar-
5.2. SNR2 isons between one-year, 4-year, and 7-year analyses due
to the duration and the parent sample selection crite-
Light-curve variability can also be quantified using
ria in Figure 5. These results indicate, although SNR2
an empirical variability metric known as the variabil-
can be a valuable preliminary screening metric, its ef-
ity signal-to-noise ratio (SNR2). This metric, as com-
fectiveness is influenced by the cadence and duration of
puted by PrepSpec, is calculated as the square root
monitoring.
of the chi-squared statistic (χ2 ) minus the degrees of
freedom (DOF), expressed as N -1, where N represents 5.3. Durbin-Watson Statistic
the
p number of data points in the light curve (SNR2 =
χ2 − DOF) and the χ2 is calculated against the mean While our findings suggest the importance of variabil-
flux. Therefore, smaller SNR2 indicates that the light ity as measured by SNR2, instances exist where a gold
curve is not variable compared to the mean flux and lag remains undetected despite relatively high SNR2 val-
higher SNR2 values indicate a light curve with greater ues. For time-series analysis software to effectively de-
intrinsic variability where the null hypothesis signifies a tect an RM lag, the light curve often requires a dis-
poor fit for a constant light curve. tinct “hook” or inflection feature in the continuum and
To enhance the efficiency of RM lag analysis, some line light curves. In this section, we investigate whether
studies have proposed pre-selecting targets based on the the presence of serial correlations in the continuum and
variability of their emission-line light curves. This ap- emission line light curve would lead to a higher success
proach aims to reduce the number of unlikely candidates in detecting a gold-lag measurement. Serial correlation
for lag detection. Early SDSS-RM studies, such as those refers to the dependence of current residuals (from a re-
by Grier et al. (2019) and Homayouni et al. (2020), em- gression model) on past residuals, indicating a patterned
ployed a threshold of SNR2 > 20 for their initial, larger behavior in the time series. A commonly used test for
parent samples. However, this practice was not univer- first-order autocorrelation, which assumes independent
sally adopted among the SDSS-RM work, potentially error terms, is the Durbin-Watson test (Durbin & Wat-
introducing bias into some early SDSS-RM results. son 1950). The Durbin–Watson test statistic is given
Both C IV and Mg II in Grier et al. (2019) and by
Homayouni et al. (2020) studies employed an SNR2 > 20
ΣTt=2 (et − et−1 )2
threshold in their parent samples, whereas the lag mea- dw = ≈ 2 − 2r, (1)
ΣTt=1 e2t
surements reported in Grier et al. (2017); Shen et al.
(2024) did not. Notably, a strong preference for this Where et is the least-square residual and r is the first
threshold exists even though it was not consistently ap- order autocorrelation coefficient. Consequently, a value
plied across all studies (Figure 5). While Shen et al. of dw ≈ 2 indicates that the first order autocorrelation
(2024) found a 70% lag detection rate for Hβ for targets coefficient r ≈ 0. If dw < 2, this is an indication for
with SNR2 > 35, a similar analysis of SNR2 in Grier positive autocorrelation r > 0; if dw > 2, then r <
et al. (2017) revealed that the majority of gold lags had 0. Therefore, we can assess the connection between the
SNR2 < 20; however, this might also be connected to autocorrelation, as indicated by the dw statistics, which
the higher cadence during the first-year of SDSS-RM can identify “hooks” in the light curves and might serve
observation (see Section 6 for a thorough discussion). as a reliable predictor of the gold-lag measurements.
To investigate the potential relationship between We illustrate the result of this comparison in Fig-
SNR2 and the success of gold-lag measurements, we ures 6 and 7 for the continuum and emission-line light
examined the lag measurements by Grier et al. (2017, curve respectively. The dw statistic, particularly for the
2019), Homayouni et al. (2020), and Shen et al. (2024). emission-line light curves, exhibits a strong preference
Our investigation revealed a direct correlation between for values below 1 within the gold lag sample (See Fig-
SNR2 values and the gold-lag detection rate for the Hβ ure 7). The trend toward dw < 1 in the gold lag sam-
sample compared to Mg II and C IV. While the detec- ple is particularly evident in the 4-year C IV lag mea-
tion rate somewhat increases with higher SNR2 for all surements of Grier et al. (2019) and the 7-year results
three emission lines, the increase was more pronounced reported by Shen et al. (2024). This suggests that a
for Hβ. Shen et al. (2024) observed a comparable trend dw < 1 might be an indicator for successful gold lag
13
90
SDSS-RM (First Year) SDSS-RM (4-year) 120 SDSS-RM (4-year)
Hβ Grier+17 (Hβ) 80
MgII Homayouni+20 (MgII) CIV Grier+19 (CIV)
50
70 100
40 60 80
MgII SNR2
CIV SNR2
Hβ SNR2
50
30 60
40
20 30 40
20
10 20
10
0
0.00 0.25 0.50 0.75 1.00 0 0.5 1.0 1.5 0 2 3 4
z z z
0.50 0.50 0.50
Success Ratio
Success Ratio
Success Ratio
0.25 0.25 0.25
0.00 0.00 0.00

0 20 40 60 0 20 40 60 80 0 25 50 75 100 125
Hβ SNR2 MgII SNR2 CIV SNR2
200 200 140

SDSS-RM (7-year) SDSS-RM (7-year) SDSS-RM (7-year)
175
Hβ Shen+24 (Hβ)
175
MgII Shen+24 (MgII) CIV Shen+24 (CIV)
120
150 150
100
125 125
MgII SNR2
CIV SNR2
Hβ SNR2
80
100 100
60
75 75
40
50 50
25 25 20
0
0.00 0.25 0.50 0.75 1.00 0 0.5 1.0 1.5 2.0 2.5 0 2 3 4
z z z
1.0
1.0 1.0
Success Ratio
Success Ratio
Success Ratio
0.5
0.5 0.5
0.0 0.0 0.0

0 50 100 150 0 50 100 150 0 25 50 75 100
Hβ SNR2 MgII SNR2 CIV SNR2
Figure 5. Emission-line variability signal-to-noise ratio (SNR2) as a function of redshift for the SDSS-RM survey. Left column
shows the results for Hβ, middle column shows the investigation for Mg II, and the right column shows the C IV results. The
early and 7-year SDSS-RM results are illustrated in reddish and blueish tones, respectively. For each work, we also include a
panel that illustrates the gold-lag success fraction as a function of SNR2 with a similar color palette. Only the lag measurements
by Grier et al. (2019) and Homayouni et al. (2020) adopted a SNR2 threshold for the initial target selection in lag measurements.
Our analysis reveals that SNR2 has a positive trend with gold lag success fraction. This is more evident in the Hβ and C IV
lines, especially in the 7-year analysis compared to the earlier lag measurements. Overall, SNR2 generally exhibits a more stable
indicator of gold-lag measurements when observed for multi-season campaigns. Nevertheless we find significant scatter for the
SNR2 and the underlying parent sample. Table 3 provides a comparison of median SNR2 values in each work.
14
detection. The Hβ lags of Grier et al. (2017) exhibit a year. Overall, the project produced a spectroscopic light
less pronounced preference for dw < 1, potentially indi- curve with 90 epochs over seven years.
cating that this threshold might be more applicable for To assess the impact of reduced cadence, our experi-
longer-term observations. Furthermore, the gold Mg II ment is to select the light curves from the gold sample of
lag measurements in Homayouni et al. (2020) display Shen et al. (2024) and decrease the epoch density for line
a wider dw distribution, which could be related to the light curves in the initial year by 40% so that the first
weaker response of the Mg II emission line to contin- year has ∼13 epochs. We maintained the same number
uum variations. The continuum dw values generally ex- of epochs as the actual observed in subsequent years
hibit a broader distribution, with most still falling below (∼13 epochs). To implement this, we first analyzed the
1.5 (see Figure 6). This indicates that the dw derived interval distribution between epochs in 2015-2017, find-
from the emission-line light curve is a better indicator ing a mean of 12 days. Employing this distribution as
of gold-lag measurements than dw computed from the a prior, we removed epochs at random until a subset of
continuum light curve. 13 remained. Epochs with shorter intervals, less than 12
Figures 6 and 7 show the distribution of the continuum days, were preferentially eliminated compared to those
light curve dw and emission line dw in comparison to the with longer cadences. This process simulated the im-
SDSS-RM dw distribution. pact of weather loss or other unforeseen events. We
Thus far, we have focused on the intrinsic properties repeated this procedure 50 times for each target, gen-
of the quasars and their light curves, investigating how erating 50 simulated light curves per quasar in the gold
these characteristics influence the success rate of RM sample. This closely resembled real observations while
lag measurements. While target selection and variabil- maintaining a reduced cadence similar to that in the
ity are crucial aspects, another important factor in RM later years.
studies is the cadence of the observations. Given the Finally, we run PyROA for the simulated line light
lag measurements from the SDSS-RM observation, we curves with their original continuum light curves and
now focus on the observational cadence and what are apply lag identification and alias removal process to get
the implications for future RM campaigns. the final lag measurement. The reason we keep the orig-
inal continuum light curves and do not remove epochs
from them is that the photometric observations for con-
6. CADENCE PROPERTIES tinuum light curves are not as expensive to make as the
6.1. Cadence Intensity in SDSS-RM spectroscopic observations and are made by telescopes
worldwide, so the continuum light curves are therefore
A main factor in RM observations is the cadence,
densely sampled. Specifically, we have 800 epochs along
as each epoch is valuable. Given the limited number
11 years and 158 epochs in 2014 (the first year of line
of epochs available due to telescope time and weather
light curves) on average, which is extremely dense with a
constraints, RM programs must optimize their cadence
cadence of 3 days per epoch approximately. We also per-
to maximize lag measurement success. Simulations by
form some tests on running PyROA for cadence reduced
Shen et al. (2015a) demonstrated the advantages of high-
light curves for both continuum and emission lines. The
cadence observations for lag detection. While their work
distribution of interval between one epoch to the next
provided valuable guidance for the SDSS-RM initial ca-
is almost uniform for each year, i.e., we just reduced
dence design, it had two limitations: 1) it was based
20% of the total epochs for continuum light curves of
on a single observation season, and 2) it used simu-
several targets selected randomly from the gold sample
lated light curves, which may not fully represent real-
and found similar lag PDFs using PYROA Therefore, for
observing conditions and lag measurement success rates.
convenience, removing epochs in the first year in the fol-
To address these shortcomings, the current work exam-
lowing texts refers to randomly reducing the density of
ines the effect of reduced cadence using actual SDSS-RM
epochs in the first year of the original emission line light
data. This is done by comparing the impact of a rela-
curve to 40% unless otherwise specified.
tively uniform cadence to the denser SDSS-RM cadence
in the initial year (2014) and a sparser cadence in subse-
quent years (2015 - 2020). The SDSS-RM project, span- 6.2. Impact of Reduced Cadence on Lag Success
ning seven years, employed a varying cadence design. In We run all the simulated light curves through the same
2014, spectroscopy occurred every four days on average, lag identification pipeline as Shen et al. (2024) and per-
totaling 32 epochs. From 2015 to 2017, two epochs were form the following statistical analyses. We use three
captured monthly, averaging 12 epochs annually. Dur- criteria to assess the difference between the original lag
ing 2018-2020, the cadence decreased to six epochs per measurements and the simulated lag measurements for
15
4 4 4
MgII Durbin-Watson
3 3 3
CIV Durbin-Watson
Hβ Durbin-Watson
2 2 2
1 1 1
0 0 0
0.0 0.5 1.0 0.5 1.0 1.5 2 3 4
z z z
1 1 1
Success Ratio
Success Ratio
Success Ratio
0 0 0
0 1 2 0 1 2 0 1 2
Hβ Durbin-Watson MgII Durbin-Watson CIV Durbin-Watson
4 4 4
Hβ SDSS-RM (7-year) MgII SDSS-RM (7-year) Hβ SDSS-RM (7-year)
MgII Durbin-Watson
3 3 3
CIV Durbin-Watson
Hβ Durbin-Watson
2 2 2
1 1 1
0 0 0
0.0 0.5 1.0 1 2 2 3 4
z z z
1.0 1.0 1.0
Success Ratio
Success Ratio
Success Ratio
0.5 0.5 0.5
0.0 0.0 0.0

0 1 2 0.0 0.5 1.0 1.5 0.0 0.5 1.0 1.5
3.0 3.0 3.0
Hβ SDSS-RM
MgII SDSS-RM
CIV SDSS-RM
7-year MgII Durbin-Watson
7-year CIV Durbin-Watson
2.5 2.5 2.5

7-year Hβ Durbin-Watson
2.0 2.0 2.0
1.5 1.5 1.5
1.0 1.0 1.0
0.5 0.5 0.5
0.00 1 2 3 0.00 1 2 3 0.00 1 2 3

First-year Hβ Durbin-Watson 4-year MgII Durbin-Watson 4-year CIV Durbin Watson
Figure 6. The Durbin-Watson statistic computed for the continuum light curves in Hβ, (left column), Mg II (middle column),
and the C IV (right column). The gold sample in each study is identified by the colored stars and the parent sample in each
case is shown with grey symbols, where the reddish color palette shows the early SDSS-RM results and blueish tones shows the
7-year results. The Durbin-Watson statistic shows the auto-correlation of the continuum light curve with a lag of one epoch.
Positive serial auto-correlation (r = 1) has a dw = 0, no serial auto-correlation has a dw = 2, and negative serial auto-correlation
has a dw = 4. The bottom row, compares the 7-year vs. the early-year SDSS-RM continuum Durbin-Watson statistics. We
see an abundance of successful lag measurements for continuum light curves that have dw < 1. While the gold-lag success rate
does not exhibit a strong correlation with continuum light curve dw, a subtle trend may be present for Hβ lags. A comparison
of median dw values in Table 3 reveals that the continuum dw of gold-lag targets is generally comparable to that of the parent
population.
16
4 4 4
MgII Durbin-Watson
3 3 3
CIV Durbin-Watson
Hβ Durbin-Watson
2 2 2
1 1 1
0 0 0
0.0 0.5 1.0 0.5 1.0 1.5 2 3 4
z z z
1 1 1
Success Ratio
Success Ratio
Success Ratio
0 0 0
0 1 2 3 0 1 2 0 1 2
4 4 4
Hβ SDSS-RM (7-year) MgII SDSS-RM (7-year) Hβ SDSS-RM (7-year)
MgII Durbin-Watson
3 3 3
CIV Durbin-Watson
Hβ Durbin-Watson
2 2 2
1 1 1
0 0 0
0.0 0.5 1.0 1 2 2 3 4
z z z
1.0 1.0 1.0
Success Ratio
Success Ratio
Success Ratio
0.5 0.5 0.5
0.0 0.0 0.0

0 1 2 0 1 2 0 1 2
3.0 3.0 3.0
Hβ SDSS-RM
MgII SDSS-RM
CIV SDSS-RM
7-year MgII Durbin-Watson
7-year CIV Durbin-Watson
2.5 2.5 2.5

7-year Hβ Durbin-Watson
2.0 2.0 2.0
1.5 1.5 1.5
1.0 1.0 1.0
0.5 0.5 0.5
0.00 1 2 3 0.00 1 2 3 0.00 1 2 3

First-year Hβ Durbin-Watson 4-year MgII Durbin-Watson 4-year CIV Durbin Watson
Figure 7. Similar to Figure 6 but for the Hβ (left column), Mg II (middle column), and C IV (right column) light curves.
Similar to Figure 6, we see that the majority of successful lag measurements correspond to light curves with dw <1. The gold-lag
success ratio is significantly correlated with emission-line dw for values less than 1. Comparative analyses indicate that the
emission line dw of gold-lag targets is consistently lower than for the overall population (see Table 3), with a more pronounced
effect for Hβ and a less significant one for Mg II.
17
1.0
Hβ 0.15 τoriginal
≥ 29 (Long)
0.8 τoriginal 0.15
< 29 (Short)
Frequency
Frequency
0.6 0.10
rmax
0.10
0.4 NLong,|eσ|≤1 = 79% NLong,|δ|≤0.34 = 90%
Nsig Original (SIG)
Original (INSIG)
0.05 NShort,|eσ|≤1 = 92% 0.05 NShort,|δ|≤0.34 = 85%
0.2 = 94%
Simulated (SIG)
Simulated (INSIG)
0.0 101 0.00 0 10 0.00 0.0 2.5 5.0
|S/N| e
Consistency Level, σ δ = (τsim − τori)/τori
1.0
Mg II 0.20 τoriginal
≥ 169 (Long)
0.8 τoriginal
0.3
0.15 < 169 (Short)
Frequency
Frequency
0.6
0.2
rmax
0.10
0.4 NLong,|eσ|≤1 = 85% NLong,|δ|≤0.34 = 94%
Nsig = 88% 0.05 NShort,|eσ|≤1 = 87% 0.1 NShort,|δ|≤0.34 = 86%
0.2
0.0 101 0.00 0 10 0.0 0.0 2.5 5.0

|S/N| e
1.0
C IV τoriginal
≥ 140 (Long)
0.8 0.20 τoriginal 0.3
< 140 (Short)
Frequency
Frequency
0.6 0.15
rmax
0.2
0.4 0.10
NLong,|eσ|≤1 = 95% NLong,|δ|≤0.34 = 95%
Nsig = 90% NShort,|eσ|≤1 = 97% 0.1 NShort,|δ|≤0.34 = 89%
0.2 0.05
0.0 101 0.00 0 10 0.0 0.0 2.5 5.0

|S/N| e
Figure 8. Summary of the simulated lag measurements from the sample of Hβ (top), Mg II (middle) and C IV (bottom) from
the gold results in (Shen et al. 2024). For each emission line, the panel on the left shows the maximum change in the statistical
criteria in the simulated light curves. Here the dotted points represent the original statistical values on rmax and lag difference
from zero, while the crosses represent the minimum statistical values among the 50 simulations for each target. The targets for
which all cadence-reduced simulations are still significant are marked in blue, while the targets for which at least one cadence-
reduced simulation is insignificant are marked in red. In each panel, Nsig reports the ratio of statistically significant simulations
to the number of all simulations. Overall, 94% of the Hβ, 88% of Mg II, and 90% of the C IV simulations are still statistically
significant. The middle panel illustrates the distribution of the consistency level between the original and simulated lag for
long and short-lag groups (separated by the median original lag of each line). Here the dashed line identify the 1σ consistency
level. The ratio of consistent lags (to within 1σ) are reported in each panel. The right panels show the distribution of relative
difference δ between the original and simulated lags divided for the short and long lag measurements. The dashed lines identify
the 34% threshold of the relative difference. The ratios of lags that have a relative difference of |δ| ≤ 0.34 are reported for the
short and long-lag samples. Overall, if we require the simulated lags are still statistically significant and consistent with original
lags in ±1σ, we can still recover 81%, 76%, and 86% of all simulations for Hβ, Mg II, and for C IV respectively.
18
lower uncertainties of original lag and upper un-

certainties of simulated lags, respectively.
• We determine the relative difference between the

median of the 50 simulated lag τsim and the orig-
inal lag τori as δ = (τsim − τori )/τori , and require
the relative difference less than 34%.
The distributions of the lag selection criteria, |S/N|,

and rmax are displayed on the left panels in Figure 8.
Most of the lag posterior samples given by PyROA will
still have a single, prominent peak in the lower-cadence
simulations, so we do not show the changes for the lag
Figure 9. Overview of the lag recovery ratio binned by red- selection criteria fpeak . While the statistical values for
shift and i-band magnitude. Here each bin reports the ratio |S/N| and rmax generally decreased in the lower-cadence
of statistically significant lag measurements that are consis- simulations, most simulated lags still met the signifi-
tent to within 1σ of the reported lag measurement in Shen cance criteria. Figure 8 highlights several targets with
et al. (2024). Overall, our light curve simulations (see Section
original |S/N| and rmax values near the selection thresh-
6.2 for detail) shows that reducing the observation cadence
by ∼ 40% during the first year would result in slightly worse old that resulted in insignificant lag measurements un-
lag recovery ratios for z < 0.8, where the Hβ emission line is der reduced cadence conditions. The simulated lags that
primary line of interest for ground-based campaigns, and of- were initially considered insignificant but later became
ten exhibits shorter lags. The result of our simulations may significant are primarily found in cases with lower over-
be conservative since they are based on the gold sample in all light curve quality. This suggests that the reduction
Shen et al. (2024), which results in ∼ 100 in Hβ, Mg II, and in measurement quality during the first season pushed
C IV emission lines.
these lags below the original selection criteria. Despite
having the lowest median observed-frame lag (31 days),
targets from the gold sample, and display the variation the Hβ sample still has the highest ratio of significant
of simulated lags in Figure 8. The three criteria are: results. This may be attributed to the generally higher
• The simulated lag is still detected with the sta- lag |S/N| and rmax values in the Hβ sample, indicating
tistically significant criteria defined in Section 3: its robustness to epoch reduction.
(1) less than half of the lag posterior samples are Our reduced-cadence simulations demonstrate that
removed by our alias removal process (as defined Hβ lags are most resilient to a reduction in number
by the fpeak parameter in Section 3); (2) The of epochs , with 94% remaining significant after a 40%
maximum cross correlation coefficient rmax > 0.4 epoch reduction in the first year of SDSS-RM. For Mg II
within ±1σ of the reported PyROA lag; (3) The and C IV, we identified 88% and 90% of significant lags,
PyROA lag is at least 2σ different from zero lag, respectively, using the criteria from Shen et al. (2024).
i.e., |S/N| > 2, and forPyROA, it always means that Overall, our simulations demonstrate that we can re-
the lag is positive at ≥ 2σ significance (Shen et al. cover ∼ 90% of significant results for all the three emis-
2024);. sion lines: Hβ, Mg II, and C IV. Hβ lags exhibit the
highest ratio of significant results, regardless of the orig-
• The simulated lag is consistent with the original inal lag value. While targets with shorter original lags
lag within 1σ uncertainties. To be more specific, might be expected to be more affected by epoch reduc-
we use consistency level σ̃ to quantitatively de- tion, we found no significant difference in the success
scribe how well the original lag and the simulated rate for detecting long or short lags.
lag are consistent within a given margin of error. We display the consistency level σ̃ for significant re-
If the simulated lag τsim is larger than the orig- sults in the middle column of Figure 8. As described
inal lag τori , σ̃ = (τsim − τori )/(σori,upper + above, the σ̃ can quantitatively describe how well the
σsim,lower ) > 0, where σori,upper , σsim,lower is the original lag and the simulated lag are consistent within
upper uncertainties of original lag and lower un- a given margin of error. We divided the sample into
certainties of simulated lags, respectively; on the long- and short-lag groups based on their median orig-
contrary, if the simulated lag τsim is smaller than inal lags: 31 days for Hβ, 155 days for Mg II, and 140
the original lag τori , σ̃ = (τsim − τori )/(σori,lower + days for C IV. Short-lag targets exhibited a higher de-
σsim,upper ) < 0, where σori,lower , σsim,upper is the gree of consistency (|σ̃| < 1) compared to long-lag tar-
19
gets across all emission lines. The most pronounced dif- 7. SUMMARY AND IMPLICATIONS FOR FUTURE
ference was seen for Hβ, with short-lag targets achieving INDUSTRIAL-SCALE SURVEYS
a 92% ± 1% consistency rate, surpassing the 79% ± 1% The Sloan Digital Sky Survey Reverberation Mapping
of long-lag targets. Mg II and C IV also exhibited sim- (SDSS-RM) project has offered a valuable dataset for in-
ilar trends, with short-lag targets outperforming long- vestigating factors that influence the successful measure-
lag targets by small margins. However, among the three ment of reliable RM lags in quasars. Previous SDSS-RM
emission lines, C IV (with the longest median original studies (Grier et al. 2017, 2019; Homayouni et al. 2020;
lag) demonstrated the highest overall ratio of simula- Shen et al. 2024) have identified 172 high-confidence lag
tions that maintained consistent results within 1σ er- measurements (the gold sample). The gold lag success
rors, while Hβ and Mg II exhibited lower ratios. The rate in SDSS-RM varies between 4% and 20% depend-
short-lag sample may exhibit higher consistency due to ing on the emission line type and survey length (see Ta-
its smaller absolute changes in lags and associated errors ble 1). To better understand the gold-lag success rate,
compared to the long-lag sample. and analysis turnaround, we have analyzed the impact
Figure 8 (right column) illustrates the distribution of of intrinsic target properties, light-curve statistics, ob-
relative differences between simulated and original lags servation cadence, and the number of observing epochs
for significant results of Hβ, Mg II, and C IV. The rel- on lag measurement. Our goal is to enhance the effi-
ative difference, δ, quantifies the change in the median ciency of lag analysis by identifying targets and light
simulated lag relative to the original. Long-lag targets curves with a higher probability of success. This will fa-
consistently exhibited a higher proportion of “similar” cilitate a more strategic approach, allowing for efficient
results (|δ| ≤ 0.34) compared to short-lag targets. For prioritization of targets and ultimately maximizing the
Hβ, this ratio was 90% ± 1% for long lags and 85% ± scientific return of surveys that are similar to SDSS-RM.
1% for short lags. Similar trends were observed for Mg II For each individual lag study, we examined the corre-
(94% vs. 86%) and C IV (95% vs. 89%). However, if lation between target properties (luminosity, equivalent
we require both statistical significance and consistency width of the targeted emission line, and redshift) and
within ±δ, the lag recovery ratios are 81% for Hβ, 76% light curve characteristics (fractional RMS variability,
for Mg II, and 86% for C IV. Targets with longer original SNR2, and Durbin-Watson statistics) with the success
lags are slightly more susceptible to a 40% reduction in of measuring gold lags. Comparison of the target prop-
cadence during the first year (∼13 epochs). erties such as luminosity λL5100 (Figure 1) and emission
To inform future cadence design, we examined the re- line equivalent width (Figure 2) does not reveal a cor-
lationship between lag recovery ratio, i-band magnitude, relation with the gold-lag measurements. We observed
and redshift, given that SDSS-RM targets are selected a significantly higher success rate (≳20%) in detecting
based on flux limits. Figure 9 presents an overview of the gold lags when the Hβ line is positioned near the cen-
lag recovery ratio binned by redshift (bin size = 0.4) and ter of the observed-frame spectroscopic range. However,
i-band magnitude (bin size = 1). Overall, we observed the impact of emission line position on detection is less
a trend of higher success ratios for brighter i-band mag- strong for Mg II due to redshift cuts and for C IV due to
nitudes and lower redshifts. Several bins, particularly potentially longer lags exceeding our baseline coverage
those near the boundary of the redshift-magnitude pa- (Figure 4).
rameter space, contain fewer than five targets and may As for the light curve characteristic, while fractional
not provide statistically meaningful results. We there- RMS variability shows no correlation with successful lag
fore excluded these bins from Figure 9. Given our results measurement, the empirical variability metric SNR2 ex-
based on seven years of observations from SDSS-RM, the hibits a positive trend. The Hβ SNR2 in the gold sample
limited number of significant lags detected at these red- of Grier et al. (2017) and Shen et al. (2024) is signifi-
shifts and magnitudes, as also indicated by mock data in cantly higher for the gold lags than those of the par-
Shen et al. (2015a), underscores the challenges of mea- ent sample. This is aligned with Shen et al. (2024)’s
suring significant lags at these cosmological distances observations on the role of variability metrics as mea-
and magnitudes. sured by SNR2 for general lag detection. Similarly, the
Our simulation results demonstrate a similar redshift gold lags measured by Grier et al. (2019) and Homay-
and magnitude dependence for the recovery ratio as ouni et al. (2020) maybe inconclusive for the impact of
found in Shen et al. (2015a), even when comparing our SNR2 as their initial samples were restricted to only
7-year observations to their 6-month mock light curves have SNR2> 20.
(Figures 8 and 9 in Shen et al. 2015a). The Durbin-Watson (dw) statistic is a reliable indica-
tor of gold-lag measurements, particularly when applied
20
to emission line light curves. While the dw statistic for results are consistent with earlier studies using shorter
the continuum light curve shows a weaker correlation, mock light curves Shen et al. (2015a). Overall, our simu-
analyzing the emission line dw reveals a stronger trend. lations indicate that a mixed cadence strategy, featuring
For Mg II, the dw statistic exhibits a weaker correlation a period of higher cadence followed by lower cadence as
with gold lag success compared to Hβ and C IV. This in SDSS-RM, provides only a marginal improvement in
is likely due to Mg II’s lower variability amplitude (Sun gold-lag measurement compared to a more uniform ca-
et al. 2015). Overall, our analysis suggests that the dw dence, which can still recover 90% of the lags.
statistic, especially when applied to emission line light
curves, is a valuable tool for efficiently filtering targets 1 The spectroscopic observations used in this study were
and predicting the likelihood of obtaining reliable lag 2 obtained by the Sloan Digital Sky Survey (SDSS) from
measurements, particularly in large datasets. 3 2015−2020. Funding for the Sloan Digital Sky Survey
Additionally, we investigate the impact of reducing 4 has been provided by the Alfred P. Sloan Foundation,
the SDSS-RM cadence on lag measurements based on 5 the Heising-Simons Foundation, the National Science
the observed quasar light curves of Shen et al. (2024). 6 Foundation, and the Participating Institutions. SDSS
We simulate multi-year light curves with 40% reduction 7 acknowledges support and resources from the Center
in cadence during the first year and compare with the 8 for High-Performance Computing at the University of
existing lag measurements for the same target. While 9 Utah. SDSS telescopes are located at Apache Point Ob-
most statistically significant lags are preserved after ca- 10 servatory, funded by the Astrophysical Research Con-
dence removal (∼90% or higher for all emission lines 11 sortium and operated by New Mexico State University,
studied: Hβ, Mg II, and C IV), a small percentage be- 12 and at Las Campanas Observatory, operated by the
come insignificant. The consistency between original 13 Carnegie Institution for Science. The SDSS web site
and simulated lags (within a margin of error) is gen- 14 is www.sdss.org.
erally higher for shorter lags compared to longer lags. 15 Y.H. was supported as an Eberly Research Fellow by
However, when we consider the lag similarity, it is worse 16 the Eberly College of Science at the Pennsylvania State
for shorter lags. This is likely due to the decreased tem- 17 University. WNB thanks NSF grants AST-2106990 and
poral resolution, which can limit the ability to capture 18 AST-2407089. LCH was supported by the National Sci-
the rapid variations associated with shorter lags. Fur- 19 ence Foundation of China (11991052, 12233001), the Na-
thermore, we find that the recovery rate is slightly lower 20 tional Key R&D Program of China (2022YFF0503401),
for fainter or higher redshift quasars (Figure 9). These 21 and the China Manned Space Project (CMS-CSST-
22 2021-A04, CMS-CSST-2021-A06).
REFERENCES
Bao, D.-W., Brotherton, M. S., Du, P., et al. 2022, arXiv Bentz, M. C., Horne, K., Barth, A. J., et al. 2010, ApJL,
e-prints, arXiv:2207.00297. 720, L46, doi: 10.1088/2041-8205/720/1/L46
https://arxiv.org/abs/2207.00297 Bentz, M. C., Denney, K. D., Grier, C. J., et al. 2013, ApJ,
Barth, A. J., Nguyen, M. L., Malkan, M. A., et al. 2011, 767, 149, doi: 10.1088/0004-637X/767/2/149
ApJ, 732, 121, doi: 10.1088/0004-637X/732/2/121 Blandford, R. D., & McKee, C. F. 1982, ApJ, 255, 419,
Barth, A. J., Bennert, V. N., Canalizo, G., et al. 2015, doi: 10.1086/159843
ApJS, 217, 26, doi: 10.1088/0067-0049/217/2/26 Blanton, M. R., Bershady, M. A., Abolfathi, B., et al. 2017,
AJ, 154, 28, doi: 10.3847/1538-3881/aa7567
Bellm, E. C., Kulkarni, S. R., Graham, M. J., et al. 2019,
Cackett, E. M., Bentz, M. C., & Kara, E. 2021, iScience,
PASP, 131, 018002, doi: 10.1088/1538-3873/aaecbe
24, 102557, doi: 10.1016/j.isci.2021.102557
Bentz, M. C., & Katz, S. 2015, Publications of the
Clavel, J., Reichert, G. A., Alloin, D., et al. 1991, ApJ, 366,
Astronomical Society of the Pacific, 127, 67,
64, doi: 10.1086/169540
doi: 10.1086/679601
Collier, S., Crenshaw, D. M., Peterson, B. M., et al. 2001,
Bentz, M. C., Williams, P. R., Street, R., et al. 2021, ApJ, ApJ, 561, 146, doi: 10.1086/323234
920, 112, doi: 10.3847/1538-4357/ac19af Collier, S. J., Horne, K., Kaspi, S., et al. 1998, ApJ, 500,
Bentz, M. C., Walsh, J. L., Barth, A. J., et al. 2008, ApJL, 162, doi: 10.1086/305720
689, L21, doi: 10.1086/595719 Dalla Bontà, E., Peterson, B. M., Bentz, M. C., et al. 2020,
—. 2009, ApJ, 705, 199, doi: 10.1088/0004-637X/705/1/199 ApJ, 903, 112, doi: 10.3847/1538-4357/abbc1c
21
de Jong, R. S., Agertz, O., Berbel, A. A., et al. 2019, The Grier, C. J., Trump, J. R., Shen, Y., et al. 2017, ApJ, 851,
Messenger, 175, 3, doi: 10.18727/0722-6691/5117 21, doi: 10.3847/1538-4357/aa98dc
De Rosa, G., Peterson, B. M., Ely, J., et al. 2015, ApJ, 806, Grier, C. J., Shen, Y., Horne, K., et al. 2019, ApJ, 887, 38,
128, doi: 10.1088/0004-637X/806/1/128 doi: 10.3847/1538-4357/ab4ea5
Denney, K. D., Horne, K., Brandt, W. N., et al. 2016a, Gunn, J. E., Siegmund, W. A., Mannery, E. J., et al. 2006,
ApJ, 833, 33, doi: 10.3847/1538-4357/833/1/33 AJ, 131, 2332, doi: 10.1086/500975
Denney, K. D., Peterson, B. M., Pogge, R. W., et al. 2010, Hemler, Z. S., Grier, C. J., Brandt, W. N., et al. 2019, ApJ,
ApJ, 721, 715, doi: 10.1088/0004-637X/721/1/715 872, 21, doi: 10.3847/1538-4357/aaf1bf
Denney, K. D., Horne, K., Shen, Y., et al. 2016b, ApJS, Homayouni, Y., Trump, J. R., Grier, C. J., et al. 2019,
224, 14, doi: 10.3847/0067-0049/224/2/14 ApJ, 880, 126, doi: 10.3847/1538-4357/ab2638
Dexter, J., Xin, S., Shen, Y., et al. 2019, arXiv e-prints, —. 2020, ApJ, 901, 55, doi: 10.3847/1538-4357/ababa9
arXiv:1906.10138. https://arxiv.org/abs/1906.10138 Homayouni, Y., De Rosa, G., Plesha, R., et al. 2023, ApJ,
Donnan, F. R., Horne, K., & Hernández Santisteban, J. V. 948, 85, doi: 10.3847/1538-4357/acc45a
2021, MNRAS, 508, 5449, doi: 10.1093/mnras/stab2832 Horne, K., De Rosa, G., Peterson, B. M., et al. 2021, ApJ,
Du, P., Hu, C., Lu, K.-X., et al. 2014, ApJ, 782, 45, 907, 76, doi: 10.3847/1538-4357/abce60
doi: 10.1088/0004-637X/782/1/45 Hu, C., Du, P., Lu, K.-X., et al. 2015, ApJ, 804, 138,
Du, P., Lu, K.-X., Zhang, Z.-X., et al. 2016a, ApJ, 825, doi: 10.1088/0004-637X/804/2/138
126, doi: 10.3847/0004-637X/825/2/126 Ivezić, Ž., Kahn, S. M., Tyson, J. A., et al. 2019, ApJ, 873,
Du, P., Lu, K.-X., Hu, C., et al. 2016b, ApJ, 820, 27, 111, doi: 10.3847/1538-4357/ab042c
doi: 10.3847/0004-637X/820/1/27 Kaiser, N., Burgett, W., Chambers, K., et al. 2010, in
Eisenstein, D. J., Weinberg, D. H., Agol, E., et al. 2011, Society of Photo-Optical Instrumentation Engineers
AJ, 142, 72, doi: 10.1088/0004-6256/142/3/72 (SPIE) Conference Series, Vol. 7733, Ground-based and
Fonseca Alvarez, G., Trump, J. R., Homayouni, Y., et al. Airborne Telescopes III, ed. L. M. Stepp, R. Gilmozzi, &
2019, arXiv e-prints, arXiv:1910.10719. H. J. Hall, 77330E, doi: 10.1117/12.859188
https://arxiv.org/abs/1910.10719 Kaspi, S., Brandt, W. N., Maoz, D., et al. 2007, ApJ, 659,
Fries, L. B., Trump, J. R., Davis, M. C., et al. 2023, ApJ, 997, doi: 10.1086/512094
948, 5, doi: 10.3847/1538-4357/acbfb7 Kaspi, S., Smith, P. S., Netzer, H., et al. 2000, ApJ, 533,
Fukugita, M., Ichikawa, T., Gunn, J. E., et al. 1996, AJ, 631, doi: 10.1086/308704
111, 1748, doi: 10.1086/117915 Kelly, B. C., Bechtold, J., & Siemiginowska, A. 2009, ApJ,
Gaskell, C. M., & Peterson, B. M. 1987, ApJS, 65, 1, 698, 895, doi: 10.1088/0004-637X/698/1/895
doi: 10.1086/191216 Kinemuchi, K., Hall, P. B., McGreer, I., et al. 2020, ApJS,
Gaskell, C. M., & Sparke, L. S. 1986, ApJ, 305, 175, 250, 10, doi: 10.3847/1538-4365/aba43f
doi: 10.1086/164238 King, A. L., Martini, P., Davis, T. M., et al. 2015, MNRAS,
GRAVITY Collaboration, Sturm, E., Dexter, J., et al. 453, 1701, doi: 10.1093/mnras/stv1718
2018, Nature, 563, 657, doi: 10.1038/s41586-018-0731-9 Kormendy, J., & Ho, L. C. 2013, arXiv e-prints,
GRAVITY Collaboration, Amorim, A., Bauböck, M., et al. arXiv:1308.6483. https://arxiv.org/abs/1308.6483
2020, A&A, 643, A154, Li, J., Shen, Y., Brandt, W. N., et al. 2019, ApJ, 884, 119,
doi: 10.1051/0004-6361/202039067 doi: 10.3847/1538-4357/ab41fb
GRAVITY Collaboration, et al. 2021, A&A, 648, A117, Li, J. I. H., Shen, Y., Ho, L. C., et al. 2021, ApJ, 906, 103,
doi: 10.1051/0004-6361/202040061 doi: 10.3847/1538-4357/abc8e6
GRAVITY Collaboration, Amorim, A., Bourdarot, G., —. 2023, ApJ, 954, 173, doi: 10.3847/1538-4357/acddda
et al. 2024, arXiv e-prints, arXiv:2401.07676, Li, Y.-R., Wang, J.-M., Hu, C., Du, P., & Bai, J.-M. 2014,
doi: 10.48550/arXiv.2401.07676 ApJL, 786, L6, doi: 10.1088/2041-8205/786/1/L6
Grier, C. J., Peterson, B. M., Pogge, R. W., et al. 2012, Lira, P., Kaspi, S., Netzer, H., et al. 2018, ApJ, 865, 56,
ApJ, 755, 60, doi: 10.1088/0004-637X/755/1/60 doi: 10.3847/1538-4357/aada45
Grier, C. J., Peterson, B. M., Horne, K., et al. 2013, ApJ, MacLeod, C. L., Ivezić, Ž., Kochanek, C. S., et al. 2010,
764, 47, doi: 10.1088/0004-637X/764/1/47 ApJ, 721, 1014, doi: 10.1088/0004-637X/721/2/1014
Grier, C. J., Brandt, W. N., Hall, P. B., et al. 2016, ApJ, MacLeod, C. L., Ivezić, Ž., Sesar, B., et al. 2012, ApJ, 753,
824, 130, doi: 10.3847/0004-637X/824/2/130 106, doi: 10.1088/0004-637X/753/2/106
22
Penton, A., Malik, U., Davis, T. M., et al. 2022, MNRAS, Sun, M., Trump, J. R., Shen, Y., et al. 2015, ApJ, 811, 42,
509, 4008, doi: 10.1093/mnras/stab3027 doi: 10.1088/0004-637X/811/1/42
Peterson, B. M. 1993, Publications of the Astronomical Swann, E., Sullivan, M., Carrick, J., et al. 2019, The
Society of the Pacific, 105, 247, doi: 10.1086/133140
Messenger, 175, 58, doi: 10.18727/0722-6691/5129
Peterson, B. M., Ferrarese, L., Gilbert, K. M., et al. 2004,
U, V., Barth, A. J., Vogler, H. A., et al. 2022, ApJ, 925, 52,
ApJ, 613, 682, doi: 10.1086/423269
Richards, G. T., Lacy, M., Storrie-Lombardi, L. J., et al. doi: 10.3847/1538-4357/ac3d26
2006, ApJS, 166, 470, doi: 10.1086/506525 Vanden Berk, D. E., Wilhite, B. C., Kron, R. G., et al.
Shen, Y., Brandt, W. N., Dawson, K. S., et al. 2015a, The 2004, ApJ, 601, 692, doi: 10.1086/380563
Astrophysical Journal Supplement Series, 216, 4, Wanders, I., Peterson, B. M., Alloin, D., et al. 1997, The
doi: 10.1088/0067-0049/216/1/4
Astrophysical Journal Supplement Series, 113, 69,
Shen, Y., Greene, J. E., Ho, L. C., et al. 2015b, The
doi: 10.1086/313054
Astrophysical Journal, 805, 96,
doi: 10.1088/0004-637X/805/2/96 Wang, S., Shen, Y., Jiang, L., et al. 2019, ApJ, 882, 4,
Shen, Y., Horne, K., Grier, C. J., et al. 2016, ApJ, 818, 30, doi: 10.3847/1538-4357/ab322b
doi: 10.3847/0004-637X/818/1/30 Yu, Z., Kochanek, C. S., Peterson, B. M., et al. 2020,
Shen, Y., Hall, P. B., Horne, K., et al. 2019a, ApJS, 241, MNRAS, 491, 6045, doi: 10.1093/mnras/stz3464
34, doi: 10.3847/1538-4365/ab074f
Yu, Z., Martini, P., Penton, A., et al. 2021, MNRAS, 507,
Shen, Y., Grier, C. J., Horne, K., et al. 2019b, ApJL, 883,
3771, doi: 10.1093/mnras/stab2244
L14, doi: 10.3847/2041-8213/ab3e0f
—. 2024, arXiv e-prints, arXiv:2305.01014, Zastrocky, T. E., Brotherton, M. S., Du, P., et al. 2024,
doi: 10.48550/arXiv.2305.01014 arXiv e-prints, arXiv:2404.07343,
Smee, S. A., Gunn, J. E., Uomoto, A., et al. 2013, AJ, 146, doi: 10.48550/arXiv.2404.07343
32, doi: 10.1088/0004-6256/146/2/32 Zu, Y., Kochanek, C. S., & Peterson, B. M. 2011, ApJ, 735,
Starkey, D. A., Horne, K., & Villforth, C. 2016, MNRAS, 80, doi: 10.1088/0004-637X/735/2/80
456, 1960, doi: 10.1093/mnras/stv2744

2412.06885v1

Uploaded by

Copyright:

Available Formats

2412.06885v1

Uploaded by

Document Information

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

2412.06885v1

Uploaded by

Copyright:

Available Formats

Draft version December 11, 2024

Typeset using LATEX twocolumn style in AASTeX631

Keywords: Active Galaxies, Quasars, Reverberation Mapping

1. INTRODUCTION a few SMBHs (GRAVITY Collaboration et al. 2018,

Median log λLλ 5100 (erg s−1)

Median log λLλ 5100 (erg s−1)

Homayouni+20 (MgII) Shen+24 (MgII) 46

Table 1. Gold Lag Summary in SDSS-RM Survey

Hβ (first-year) Grier et al. (2017) 222 26 12%

150 150 150

MgII REW (Å)

00.0 0.5 1.0 0 0.5 1.0 1.5 0 2 3 4

0.0 0.0 0.0

200 200 200

150 150 150

CIV REW (Å)

100 100 100

00.0 0.5 1.0 0 1 2 0 2 3 4

0.4 Shen+24 (Hβ) 0.4 Shen+24 (MgII) 0.4 Shen+24 (CIV)

0.2 0.2 0.2

0.0 0.0 0.0

Shen+24 (Hβ) Shen+24 (MgII) Shen+24 (CIV)

150 150 150

100 100 100

00 100 200 00 100 200 00 100 200

Table 2. Comparative Analysis of Sample Properties: Gold and Parent

Emission Line λL5100 REW

0.4 0.4 0.4

0.3 10 0.3 0.3

Table 3. Comparative Analysis of Light Curve Properties: Gold and Parent

Emission Line Frac. RMS Var. SNR2 Con. dw Line dw

Parent Gold Parent Gold Parent Gold Parent Gold

MgII Fractional RMS Variability

CIV Fractional RMS Variability

0.6 0.6 0.6

0.4 0.4 0.4

0.2 0.2 0.2

0.0 0.0 0.0

1.0 1.0 1.0

CIV Fractional RMS Variability

0.8 0.8 0.8

0.6 0.6 0.6

0.4 0.4 0.4

0.2 0.2 0.2

0.5 0.5 0.5

0.0 0.0 0.0

Comparison Comparison Comparison

7-year CIV Fractional RMS Variability

Grier+17 (Hβ) Homayouni+20 (MgII) Grier+19 (CIV)

0.6 0.6 0.6

0.4 0.4 0.4

0.2 0.2 0.2

0.0 0.00.0 0.00.0

0.00 0.00 0.00

200 200 140

0.0 0.0 0.0

Shen+24 (Hβ) Shen+24 (MgII) Shen+24 (CIV)

0.5 0.5 0.5

0.0 0.0 0.0

7-year CIV Durbin-Watson