Changing-look Active Galactic Nuclei from the Dark Energy Spectroscopic Instrument. I. Sample from the Early Data

Changing-look active galactic nuclei (CL AGNs) can be generally confirmed by the emergence (turn-on) or disappearance (turn-off) of broad emission lines (BELs), associated with a transient timescale (about 100 ∼ 5000 days) that is much shorter than predicted by traditional accretion disk models. We carry out a systematic CL AGN search by crossmatching the spectra coming from the Dark Energy Spectroscopic Instrument and the Sloan Digital Sky Survey. Following previous studies, we identify CL AGNs based on Hα, Hβ, and Mg ii at z ≤ 0.75 and Mg ii, C iii], and C iv at z > 0.75. We present 56 CL AGNs based on visual inspection and three selection criteria, including 2 Hα, 34 Hβ, 9 Mg ii, 18 C iii], and 1 C iv CL AGN. Eight cases show simultaneous appearances/disappearances of two BELs. We also present 44 CL AGN candidates with significant flux variation of BELs, but remaining strong broad components. In the confirmed CL AGNs, 10 cases show additional CL candidate features for different lines. In this paper, we find: (1) a 24:32 ratio of turn-on to turn-off CL AGNs; (2) an upper-limit transition timescale ranging from 330 to 5762 days in the rest frame; and (3) the majority of CL AGNs follow the bluer-when-brighter trend. Our results greatly increase the current CL census (∼30%) and would be conducive to exploring the underlying physical mechanism.


INTRODUCTION
In the unification paradigm, different types of AGN or quasars 1 , are classified by their orientation relative to the line-of-sight, and present significant flux variation (about 10%-30%) across the entire electromagnetic spectrum on monthly to annual timescales (Antonucci 1993;Vanden Berk et al. 2004).The Broad Line Region (BLR) is located close to the central supermassive black hole (a few light-days to lightmonths distance).Ionized gas of the BLR emits Broad Emission Lines (BELs) with a smaller flux variation amplitude than the continuum (Kaspi et al. 2000;Du & Wang 2019).These BELs (in particular, the Balmer lines) have been found to respond to continuum variations with a time delay.This time delay has been used, in a technique known as reverberation mapping, to study the geometry and kinematics of BLR and measure the central black hole mass (Blandford & Mc-Kee 1982;Peterson 1993;Bentz et al. 2009;Ho & Kim 2014;Li et al. 2018;Lu et al. 2021;Bao et al. 2022).The variation of BEL features is therefore an effective tool for studying the formation of the BLR and the evolution of AGN.However, Mg II and C IV are less responsive to the continuum variation than Balmer lines because of their intrinsic properties, such as weak response, the extended BLR size, or outflows (Richards et al. 2011;Sun et al. 2018;MacLeod et al. 2019;Yu et al. 2021).
The changing-look (CL) phenomenon was originally used to characterize X-ray detected AGN that change between Compton-thick and Compton-thin (Matt et al. 2003;Temple et al. 2022).In the optical band, CL is a surprising and unusual event in which BELs in AGN spectra appear or disappear (from Type 1 to Type 2 or vice versa) within just a few months or years.In low-redshift AGN (z < 0.1), CL could also refer to the transition between intermediate types (such as Type 1.2, Type 1.5, Type 1.8, and Type 1.9) that is determined by the flux ratio between Hβ and [O III]2 (Osterbrock 1981;Cohen et al. 1986;Winkler 1992;Denney et al. 2014;Li et al. 2022b).The discovery of the first CL quasar, SDSS J015957.64+003310.5, by LaMassa et al. (2015) which underwent a dramatic change from Type 1 to Type 1.9 in just ten years, poses a challenge to our understanding of the AGN unified model and the widely accepted viscous heating timescale (hundreds of years) of a steady accretion disk (Shakura & Sunyaev 1973;Cackett et al. 2007;Runnoe et al. 2016;Tie & Kochanek 2018;Guo et al. 2022).The study of CL AGN evolution can also shed light on the growth and feedback processes of supermassive black holes and the impact of AGN on their host galaxies (Ho 2005;Wang & Zhang 2007;Zhuang & Ho 2019;Jin et al. 2022).
Over a hundred CL AGN have been discovered based on their Balmer line profile transitions (MacLeod et al. 2016;Gezari et al. 2017;Sheng et al. 2017;Frederick et al. 2019;Hon et al. 2022;Green et al. 2022;Wang et al. 2022;Zeltyn et al. 2022;Yang et al. 2023), and five CL AGN have also been identified based on transitions in Mg II, C III], and C IV (Guo et al. 2019;Ross et al. 2020).However, the nature and frequency of CL AGN are still not well understood, and many questions remain unanswered.For instance, the occurrence rate of CL events in AGN and the relationship between CL AGN and normal AGN are unclear.Additionally, the mechanisms responsible for these changes are yet to be fully understood.Three commonly proposed possible causes are 1) changes in the central gas density due to variation of energy radiation intensity in nuclear of the central point source, or accelerating outflows (Shapovalova et al. 2010;LaMassa et al. 2015;Ricci & Trakhtenbrot 2022); 2) rapid increase or decrease of gas density and accretion rate in the compact region originating from tidal disruption events (TDE; Blanchard et al. 2017;Li et al. 2022a; 3) accretion rate changes caused by BLR evolution, accretion disk instability, or temporary accretion events (Esin et al. 1997;Elitzur & Ho 2009;Dexter & Agol 2011;Elitzur et al. 2014;MacLeod et al. 2019;Śniegowska & Czerny 2019).CL AGN may be attributed to various mechanisms or influenced by multiple physical processes.
To better understand the physical mechanism behind the CL phenomenon, it is essential and pressing to conduct further spectral monitoring of previously discovered objects and to additionally search for new CL AGN.There are several effective methods to hunt for CL AGN, including 1) crossmatching AGN spectra in multiple-epoch large-area spectroscopic projects, such as those performed by MacLeod et al. 2016;Yang et al. 2018;Green et al. 2022;Wang et al. 2022; and 2) identifying potential candidates through follow-up spectroscopic observations, by detecting extremely unusual variability in optical or mid-IR light curves, as seen in studies by Sheng et al. 2017;MacLeod et al. 2019;Graham et al. 2020.The Dark Energy Spectroscopic Instrument (DESI) offers an excellent opportunity to hunt for CL AGN, as it boasts a large field of view, high spectral resolution, and high data generation efficiency (as described in detail in Section 2).Another advantage of DESI is the high sensitivity allowing for the identification of lower-luminosity accretion events.
We aim to explore the physical mechanism behind the CL AGN and improve the completeness of the current sample.
To achieve this, we will utilize the spectroscopic data from DESI and the Sloan Digital Sky Survey (SDSS).The DESI project will provide nearly three million quasar spectra in the next five years, while the SDSS project has already identified 750,414 quasars.By cross-matching the AGN spectra from these two projects, we will be able to compile a large sample of CL AGN and study their behaviors.Previous studies have primarily focused on Balmer line (Hα and Hβ) CL AGN at redshifts z < 0.7 (MacLeod et al. 2016;Yang et al. 2018;MacLeod et al. 2019;Graham et al. 2020;Green et al. 2022).Since high ionization BELs are thought to connect the inner region of the accretion disk and the BLR (Guo et al. 2019;Ross et al. 2020), we aim to study CL AGN for several major BELs at all redshifts, including Hα, Hβ, Mg II, C III], and C IV. Therefore, we will divide the sample into two parts based on redshift (z ≤ 0.75 and z > 0.75) to search for CL AGN in both the rest-frame ultraviolet and optical wavelengths.The results of this study will provide crucial insights into the properties of CL AGN and contribute to our understanding of the AGN population.
The paper is structured as follows.In Section 2, we describe the data from the DESI and SDSS projects.Section 3 outlines the process for selecting our target sample, including the use of both visual inspection and spectral variability definitions to identify CL objects.In Section 4, we present our findings on the five broad emission line CL behaviors, and upper limit timescales, and discuss potential physical origins.The paper concludes with a summary in Section 5. Throughout, we adopt a ΛCDM cosmology with H 0 = 67 km s −1 Mpc −1 , Ω Λ = 0.68, and Ω m = 0.32 as reported by Planck Collaboration et al. 2020.
2. DATA 2.1.DESI DESI3 uses the NOIRLab 4m Mayall telescope (8 deg 2 field view) at Kitt Peak with the aim to constrain the nature of dark energy and probe cosmological distances through the baryon acoustic oscillation (BAO) technique (Levi et al. 2013;DESI Collaboration et al. 2016a,b;Silber et al. 2023;Miller et al. 2023;Schlafly et al. 2023;Raichoor et al. 2023;Kirkby et al. 2023).DESI is carrying out the largest ever multiobject and high-efficiency spectral survey (5000 spectra  in a single exposure) and plans to measure 40 million galaxies and quasars within five years (DESI Collaboration et al. 2016a;Zhou et al. 2020;DESI Collaboration et al. 2022;Raichoor et al. 2023;Allende Prieto et al. 2020).The DESI program will accumulate about three million quasar spectra to measure large-scale structures in the main survey (Lan et al. 2023;Hahn et al. 2022;Ruiz-Macias et al. 2020;Yèche et al. 2020).Since the limiting magnitude of DESI in the r-band reaches about 23 mag, DESI has the unique opportunity to discover fainter and higher redshift quasars than any previous survey, quadrupling the number of quasars discovered by SDSS (Zou et al. 2018;Chaussidon et al. 2022;Alexander et al. 2023).Besides the strengths in quantity and depth, three optical channels (blue: 3600-5900 Å, green: 5660-7220 Å, and red: 7470-9800 Å) also provide an excellent spectral resolution (blue: R ∼ 2100, green: R ∼ 3200, and red: R ∼ 4100; Abareshi et al. 2022).Guy et al. (2022) describe the spectroscopic data processing pipeline in detail and references therein present the target selection and validation (Alexander et al. 2023;Lan et al. 2023;Raichoor et al. 2023;Zhou et al. 2022;Myers et al. 2023;Cooper et al. 2023).Before the start of the main survey, data from the Survey Validation (SV) was visually inspected to check the spectroscopic quality and redshift reliability (DESI Collaboration et al. 2023a,b), improve the standard DESI spectroscopic pipelines, such as the "Redrock" and the "afterburner" codes, and reduce the number of misclassified quasars (Alexander et al. 2023).
With a large sample of quasars and high-quality spectra, DESI offers unprecedented advantages for identifying CL AGN.Since DESI has the advantage of observing fainter AGN than SDSS, more turn-off CL AGN are expected to be discovered.In this project, we used 347,201 quasar spectra and 2,955,168 galaxy spectra reduced by the DESI pipeline (based on the SPECTYPE from "Redrock") with ZWARN = 0 (Brodzeller et al. 2023;Bailey et al. 2023), indicating reliable spectroscopic redshift measurements, which contains SV (named as "fuji") and the first 2 months of Year 1 data (named as "guadalupe", DESI Collaboration et al. 2023b).The data of "fuji" is published as part of the Early Data Release and "guadalupe" will be published at the same time as Data Release 1 (DR1; Alexander et al. 2023;Lan et al. 2023;Raichoor et al. 2023;DESI Collaboration et al. 2023a).

SDSS/BOSS/eBOSS
In addition to the DESI data, we used the SDSS 4 spectroscopic data, which covers a large region of the sky and has millions of spectra of galaxies and quasars (Gunn et al. 2006).The SDSS spectroscopic database provides additional highquality data for the comparison and validation of the DESI redshift measurement.The SDSS is carried out on a 2.5m Sloan telescope at Apache Point Observatory and has provided a large database of quasar and galaxy spectra through four-stage projects (SDSS-I, SDSS-II, SDSS-III/BOSS, and 4 https://www.sdss.org/SDSS-IV/eBOSS; Abazajian et al. 2009;Alam et al. 2017;Lyke et al. 2020).The spectra have a wavelength coverage of 3900-9100 Å or 3600-10400 Å and a resolution of R ∼ 2000 for quasars and galaxies (Eisenstein et al. 2011;Smee et al. 2013).All the SDSS spectra are reduced with SDSS I/II and BOSS data pipelines (Stoughton et al. 2002;Bolton et al. 2012).In this study, we have used 750,414 quasar spectra from the SDSS Data Release 16 Quasar (DR16Q) catalog released by Lyke et al. (2020) and 4,930,400 galaxy spectra from the DR16 (Ahumada et al. 2020)

DESI Legacy Survey
Motivated by the target selection for DESI, the DESI Legacy Surveys5 team utilized a deep and large area image survey (Dey et al. 2019).The survey comprised three projects: the Dark Energy Camera Legacy Survey, the Beijing-Arizona Sky Survey, and the Mayall z-band Legacy Survey, which covered approximately 14,000 square degrees of extragalactic sky, including 9,900 deg 2 in the NGC and 4,400 deg 2 in the SGC (Flaugher et al. 2015;Zhou et al. 2022;Schlegel et al. 2023).The survey was conducted using three optical/infrared bands, reaching approximate AB magnitudes of g = 24.0,r = 23.4,and z = 22.5 (Dey et al. 2019).We used the DESI Legacy Survey photometric magnitude to describe the distribution of the final CL AGN sample.

Data Preprocessing
To facilitate the subsequent analyses, including spectrum inspection, defining the change of flux threshold, and sample selection, we corrected the spectra for galactic extinction by using the galactic extinction curve of Fitzpatrick (1999) by assuming R V = 3.1.After that we shift the spectrum to the rest frame.
Since the spectra of SDSS and DESI have different wavelength coverages and resolutions, the flux variation can not be obtained directly by subtraction with two original spectra.Therefore, we rebin both spectra into the same wavelength grid for flux and its variance (2 Å per pixel) to be able to reliably compare and subtract the SDSS and DESI spectra.

SAMPLE SELECTION
Although DESI has more than one spectrum for some targets, we only use one epoch of DESI for each target and remove any duplicates.This approach does not affect the results of our study as the time interval between the different epochs of DESI spectra is within one year, which is usually shorter than the time scale of most CL objects (for more details, see Section 4.2).
The sample selection process for CL AGN in this study involves four steps: 1) build AGN parent sample that are both observed by DESI and SDSS; 2) first stage of visual

Parent Sample
We used all the spectra from the DESI catalog and SDSS DR16 catalog to systematically search the CL AGN or candidates.To obtain a parent sample of AGN with two epochs of spectra, we cross-matched the DESI spectra catalog with the SDSS catalog using a 1 ′′ separation, including both quasars and galaxies.After that, we obtained SDSS spectra of the matched sample through the Science Archive Server (SAS) 6 .We note here that about 4% of spectra are missing through SAS downloading, which is possibly due to the data quality.
Although the spectral classification pipelines of SDSS and DESI perform very efficiently, those objects with indistinct BEL characteristics or low-quality spectra could be leading to a misclassification of the spectral type (Alexander et al. 2023;Lan et al. 2023), especially for CL AGN.As pointed out by many previous works (MacLeod et al. 2016;Yang et al. 2018;Green et al. 2022), the CL AGN in a low state might be classified as galaxies because the BELs and blue continuum are no longer the dominant features.Therefore, we also included cross-matched results between AGN with galaxies.Specifically, we divided the sample into three groups: group I (DESI quasar & DR16 quasar), group II (DESI galaxy & DR16 quasar), and group III (DESI quasar & DR16 galaxy).Finally, we built up a parent sample containing 82,653 matched AGN spectrum pairs (at least one source in the cross-match is marked as AGN). Figure 1 displays the redshift distribution of our parent sample, where the maximum redshift reaches z ∼ 6.Since previous works focus on the Hα and Hβ CL phenomenon (MacLeod et al. 2016(MacLeod et al. , 2019;;Green et al. 2022), we divided the DESI spectra into two sub-samples according to the redshift to search for: 1) CL AGN defined by Hα, Hβ, and Mg II at z ≤ 0.75; 2) CL AGN defined by Mg II, C III] and C IV at z > 0.75.Another reason why we set z = 0.75 as the boundary between the two sub-samples is that we can use [O III] which can be observed in SDSS spectra out to z ∼ 0.8 to test the variability definition and estimate the final CL sample purity at z > 0.75, which required the inclusion of both Hβ and [O III] in the spectra (for more details, see Section 3.4).The number of AGN in each of the three groups at z ≤ 0.75 and z > 0.75 are given in Table 1 and Table 2, respectively.One might expect that a CL event would always have one epoch classified as an AGN and another as a galaxy (groups II and III).However, we find that most often both are classified as AGN (group I).This is primarily because most of the CL AGN we discover are in the period of transition, so still retain some AGN lines.

Visual Inspection
Previous studies effectively diagnose CL AGN based on VI of large area sky surveys (MacLeod et al. 2016;Yang et al. 2018;MacLeod et al. 2019;Green et al. 2022;Wang et al. 2022).Since the appearance or disappearance of BEL may be within a continuous process of spectral enhancement or weakening, it is a challenge to quantitatively describe the CL behavior with only two randomly sampled spectra.As mentioned by Green et al. (2022), a standard CL AGN target selection process starts with a VI to discard those spectra that have poor quality, large measurement errors, or wrong redshift identifications.
In this project, we also carry out VI of the parent sample for three purposes.Firstly, we need to remove those fake CL AGN whose behavior might be caused by spectrum defects and/or disagreement between the SDSS and DESI redshifts.For example, the absorption associated with the C IV line (for example, Broad Absorption Line AGN (Rogerson et al. 2018)) would impact the accuracy of the total BEL flux.Secondly, since the wavelength region of Hα contains strong narrow emission lines (narrow Hα and [N II]), the variation of BEL flux determined by the integration method could be affecting the selection.We inspect all the Hα CL candidates to confirm they are indeed CL AGN in the selection program (see detail in Section 4).
Thirdly, the final spectral variability definition that we apply is the automatic checker, which was established based on the results from the VI and run on all the parent samples (see Section 3.3).In conclusion, the VI check is a crucial step in searching for CL AGN and candidates in large sky surveys, as it helps to remove false positive results and establish a reliable spectral variability definition.

Spectral Variability Definition (Selection Criteria)
Before defining the variability of the BEL, we require that the median S/N per pixel is at least greater than one for both the DESI and SDSS spectrum.In previous studies (e.g.MacLeod et al. 2016;Yang et al. 2018;Graham et al. 2020;Green et al. 2022;Temple et al. 2022), a variety of criteria were chosen to describe the CL behavior of the BELs.Based on these previous studies, we adopt three spectrum parameters (N σ , R, and F σ,dim ) to accurately describe the CL phenomenon.
N σ is defined as the significance in the variation of the BEL maximum flux and is a widely used definition applied for Hβ CL AGN (MacLeod et al. 2019;Green et al. 2022), which is the flux deviation between the dim spectrum and bright spectrum: where f and σ are the spectral flux and variance respectively in erg cm −2 s −1 Å−1 .Green et al. (2022) smoothed the N σ (Hβ) array, subtracted the N σ (4750 Å) by the flux deviation array, and found the maximum relative value of The AGN with N ′ σ (Hβ) > 3 are considered as a CL object (see Green et al. 2022 Section 3.5 for details).Although this determination is effective at determining the "BEL disappeared or appeared" behavior, they also noticed that their selection criteria were very sensitive to minor BEL changes if the spectrum S/N is sufficiently high, which can be seen in several of the CL objects discovered in MacLeod et al. (2019) and Green et al. (2022).
In this paper, we aim to define a variability definition that can recreate our VI results and is also less biased by spectrum S/N.We use the maximum value of the smoothed N σ > 3 objects without the subtraction of the N σ (4750 Å) as the first determination of BEL variation to limit the overall variation of both continuum and BELs.
The second parameter (R) is defined as the ratio of the integrated BEL flux in the high-state spectrum to the integrated BEL flux in the low-state spectrum, which is used to constrain the total flux change of BELs.As mentioned and inspired by MacLeod et al. (2019), we further adopt the R value to constrain the overall BEL flux change, which is defined as: where F bright or F dim is the BEL total flux.Integration and spectral fitting are two widely used tools to obtain the BEL flux in the field of reverberation mapping (Peterson 1993;Hu et al. 2020).The integration method used in this work provides a quick and easy way to measure the broad emission line flux, which is suitable for the massive AGN dataset released by DESI.By subtracting the continuum as a straight line defined by two continuum windows, the BEL flux is measured by a simple integration method (see detail in  2020)).The BEL integration windows are carefully adjusted to ensure the best match with the first stage of VI results.The information on the integration windows is listed in Table 3.
Based on the criteria defined in Winkler (1992), An AGN would be classified as different sub-types if the Hβ and [O III] flux ratio change is larger than 1.5, such as from Type 1.2 to Type 1.5 or from Type 1.5 to Type 1.8 (see definition in Winkler (1992) for detail), which we would refer to a CL candidate in our sample.Based on Hβ, we adopt R > 1.5 as the CL candidates selection criterion for all BELs.
The last parameter F σ,dim represents the significance of the weak BEL, which is calculated from the dim spectrum7 : Since the narrow component emission lines (such as narrow Hβ, [N II]) could exist in the dim spectra to increase F σ,dim value, we take a different F σ,dim value for Balmer lines and other lines.Another reason for the lower F σ,dim value for other lines is mainly due to a lower S/N for high redshift AGN.
When the S/N is high enough, the AGN with a tiny broad component may be categorized into CL candidates.Compared with the weak component, we care more about the flux variation ratio.Thus, in the case of R > 1.5, we restrict F σ,dim < 5 for Balmer lines (F σ,dim < 3 for other lines) as the criterion for CL AGN.As R increases, we lower the limit of F σ,dim and use a straight line to divide CL AGN and candidates with F σ,dim = 3.33R for Balmer lines or F σ,dim = 2.0R for other lines, which is plotted in Figure 2.
By combining these three parameters, we can accurately describe the CL behavior of the BELs.As mentioned before, our CL AGN are mainly for the transition between type 1.x to type 1.9, while CL candidates are mainly an intermediate transition, such as similar to from type 1.2 (1.5) to 1.5 (1.8) for Hβ.The final variability definitions of these parameter thresholds are also established based on the results from the VI check and the comparison with previous studies.The final selection criteria for CL candidates and AGN are: • CL candidates: Max(N σ ) > 3, R > 1.5; • CL AGN for Hα and It is important to keep in mind that these parameter thresholds are tentative and may need to be refined or adjusted based on future large datasets.This selection method is expected to be less biased by the spectral S/N and provide a robust and efficient way to identify CL AGN in the DESI project.We also note that spectral decomposition may result in identifying weak broad components even though the BEL disappeared in VI results.Therefore, regardless of whether the BEL is disappearing or not, CL candidates and CL AGN may be incorrectly identified due to the coincidence of spectral sampling or weak BEL flux covered by the host galaxy (this will be discussed in Section 4.3).
In Figure 2 you can see there is a clear correlation between these three parameters, revealed by plotting two parameters and colour-coding the third.The population of AGN do show a continuum of these parameters, with the CL AGN the most extreme end of the distribution.However, these plots make it clear that the most important parameter for distinguishing CL AGN is the R parameter, with CL AGN sitting in a very sparsely populated region of R values.The log scale in these plots shows that the R values of CL AGN are typically orders of magnitude higher than most AGN.On the other hand CL AGN are in the same Max(N σ ) range as about a third of the AGN in our sample, and are in the same F σ,dim range as about one fifth of the sample.

[O III]-based Calibration
Several previous studies carried out a further flux calibration based on either an [O III] (5007 Å) calibration or using a photometric light curve to exclude fiber position and/or aperture issues introduced from the program selection (Yang et al. 2018;MacLeod et al. 2019;Green et al. 2022).The [O III] flux should remain relatively unchanged since the [O III] emission line probes the region reaching hundreds of parsecs to several kiloparsecs away from the central ionization source and originates from the narrow line region and host galaxy.The photometric light curve also provides a criterion to check if there is a flux calibration problem between the DESI and SDSS spectra.However, as shown in Table 4, we are unable to obtain a light curve for every CL AGN and candidates, since the majority of DESI objects are fainter than the limiting magnitude of many time-domain surveys (such as ZTF, which has a limit magnitude at 20.6 (Masci et al. 2019), but DESI can reach about 23 mag in the r-band).
To match the resolution of the two datasets, we use a Gaussian kernel to convolve DESI spectra to the lower resolution of the SDSS spectra (R ∼ 2000).Following Du et al. (2018)  and 10 C IV CL objects.While J221044.76+245958.0 is classified as CL candidates in the selection criteria, we moved it from CL candidates into CL AGN by VI because the significance of their broad components are increased by narrow lines.Figure 3 shows five examples of Hα, Hβ, Mg II, C III], and C IV CL AGN respectively, where the difference spectrum is the subtraction of the dim spectrum from the bright spectrum.Table 4 summarizes the properties of the 130 CL AGN, while the full CL AGN figures would be availabled in its entirety in machinereadable form.
To further study the whole process of CL AGN, we also search for CL AGN candidates which are in the process of changing or about to happen.In the 130 CL AGN we have discovered, 42 AGN have candidate additional BEL variations (listed in Table 5).We also provide another 91 CL candidates for five BELs as shown in Table 6.The whole CL candidate sample contains 133 AGN, including 8 Hα, 20Hβ, 45 Mg II, 33 C III] and 43 C IV CL candidates.To highlight the difference between the confirmed CL AGN and the candidates, Figure 7 provides five selected examples of CL candidate spectra.For the sake of illustration, Figure 1 shows the distribution of the redshift and r-band magnitude ( DESI legacy suvery) for CL AGN or candidates are quite similar, where the highest redshift of CL AGN reaches z = 3.56 (see Figure 6).

Time Scale and Transition State
In Figure 8, we plot the time scale of both the 130 confirmed CL AGN and the 134 CL candidates between the two observations in the observed-frame and rest-frame.We note that the number of CL AGN that show different BEL behavior from the same source are counted repeatedly.For instance, a case with varying Hβ and Mg II will be once in the Mg II sample (green histogram) and once in the Hβ sample (blue histogram).Zeltyn et al. (2022) reported a time scale for the Hβ variation observed in J162829.18+432948.5 of only a few months (73 days), whereas we probe a time scale of up to 5643 days between the SDSS and DESI ob- servations.The time scale represents the upper limit of the transition time scale and ranges from 244 to 5762 days.Figure 8 demonstrates that CL AGN with transition times scales < 1000 days are dominated by the C III] and C IV BELs in the rest-frame.By contrast, Hα and Hβ usually require a longer time to exhibit the CL phenomenon.Since the redshift of our CL AGN ranges from 0.12 ∼ 3.56, we also present the results of the observed-frame to investigate selection effect from the time dilation.While the observed-frame histogram suggests that some differences may be attributed to a selection effect, systematic differences between different samples persist.One reason may be due to the selection bias due to a small sample size, which requires further investigation with a larger sample.
In our sample of 130 CL AGN, 95 AGN exhibit turn-on behavior while 35 AGN exhibit turn-off behavior, as listed in Table 4.Among the 46 AGN with redshift z ≤ 0.75, there are 20 turn-on and 26 turn-off CL AGN.As a comparison, only 9 objects exhibit the turn-off behavior while 75 objects show the turn-on pattern for the high redshift sample.The proportion of turn-off and turn-on for Hβ CL AGN is consistent with previous work (MacLeod et al. 2016;Yang et al. 2018;MacLeod et al. 2019;Green et al. 2022;Wang et al. 2022).The discrepancy in the proportion of turn-on and turn-off CL AGN presented in different redshift samples could be a selection effect or may indicate that the time scale for turn-on and turn-off are different.As shown in the Figure 9, we find that the on-off time scales are different for different emission lines.However, due to the limited number of samples, whether this is related to the proportion of turn-on and turn-off CL AGN still needs further more complete sample verification.
We find that the median time scale in the rest-frame for seven Hβ and Mg II simultaneous CL AGN (3972 days) is conspicuous larger than time scale for single Hβ CL AGN (2692 days) or Mg II CL AGN (1044 days), which is consistent with Guo et al. 2019Guo et al. , 2020 who provided a CL sequence that CL of Mg II occurs after Hβ.However, for those CL .The histograms of the upper limit of the transition time scale for CL AGN and CL candidates in observed-frame (left panel) and rest-frame (right panel).The dash lines represent the median value of three samples.While it seems like C III] and C IV have lower rest-frame time scales than the other lines, the observed-frame histogram shows that that is partially due to a selection effect, since time-dilation has made it impossible from our sample to detect carbon rest-frame lags longer than 3500 days.
AGN with C III] and Mg II transition at the same time, the corresponding time scale (869 days) is smaller than or close to the time scale of a single emission line transition (C III]: 970 days).Whether there is an analogously evolutionary sequence or CL sequence between C III] and Mg II would be worthy of careful study in the future, which may unravel the physical origin of CL AGN.

Physical Origin
Despite the limitations in information about the continuum luminosity and power-law index, it is evident that the majority of CL AGN from Figure 3, particularly for Hα and Hβ, display the characteristic of being bluer when brighter (the optical and ultraviolet emission from the nucleus of an AGN becomes bluer when the overall brightness of the AGN increases), which is in line with many previous studies (LaMassa et al. 2015;Runnoe et al. 2016;MacLeod et al. 2016;Gezari et al. 2017;Yang et al. 2018;Graham et al. 2020;Green et al. 2022).From a quantitative standpoint, MacLeod et al. ( 2019) and Green et al. (2022) presented two separate CL AGN samples obtained through the method of spectral fitting, and both find a strong correlation between the change in Hβ flux and the variation in continuum flux (refer to Figure 3 2022) for more details).This close relationship between the continuum and BEL variation suggests that the radiation from the accretion disk continues to drive the evolution of the broad line region even during the changing state.This phenomenon also implies that there is a rapid exchange of both material and energy occurring between the accretion disk and the broad line region, indicating that the origin of CL AGN is more likely to come from changes in the accretion rate.
Despite this, there are still some uncertainties in the identification of CL AGN.According to MacLeod et al. 2016 andGreen et al. 2022, some CL AGN exhibit back-and-forth variability, where the weak Hβ broad component disappears in one epoch but reappears in another.This suggests that the discovery of CL AGN is like a random sampling from a pool with constantly changing BELs.
One possible explanation for the CL behavior is that the AGN is undergoing a transition from a low-to-high accretion rate state or vice versa.In the low-accretion rate state, the emission from the central region is relatively weak, while in the high-accretion rate state, the emission is much stronger.The transition between these two states can be triggered by various physical processes, such as the instability of the accretion disk or changes in the mass-supply rate.Elitzur & Ho (2009) proposed that AGN may have a critical luminosity that switches the BEL appearance and disappearance.This model is also supported by the discovered CL AGN, which shows a low accretion rate and low Eddington ratio (usually < 0.1) in the dim spectrum.Given that the radiation regions of C III] and C IV are relatively small, another potential scenario that could describe the C III] and C IV CL behavior is the association with an outflow Given that the majority of CL AGN display the "bluer when brighter" trend, another potential explanation could be a change in obscuration of the Broad Line Region, either due to dust clouds that move in and out of our line-ofsight or powerful winds that "blow-out" the dust surrounding the nucleus.However, the former scenario has not been supported by polarization observations, time scale calculations, and mid-infrared features for the majority CL AGN (LaMassa et al. 2015;MacLeod et al. 2016;Sheng et al. 2017).Many of the turn-on CL AGN exhibit a wavelengthdependent gain in continuum flux, appearing to change from a "red quasar" (e.g., Klindt et al. 2019;Fawcett et al. 2020) to a typical blue quasar (see Figure 3).This could indicate that dust is being removed from these systems, which may be evidence that these AGN are undergoing a "blow-out" phase, in which powerful outflows clear out the surrounding gas/dust (Glikman et al. 2017;Stacey et al. 2022;Fawcett et al. 2022).Using spectroscopy over a 19-year period, Yi et al. (2022) discovered a turn-on CL AGN that hosted powerful outflows, concluding that the quasar was shedding a surrounding dust cocoon, transitioning to a blue quasar.However, it is still unclear whether changes in the accretion disc could account for the change in continuum flux and also whether the difference in the dust extinction values over a 10-20-year timescale is consistent with a blow-out phase.Several possible models have also been proposed to explain the CL phenomenon, including the cooling front model caused by a change in the magnetic field (Ross et al. 2020), close binary black hole model (Wang & Bon 2020), magnetic field-coupled accretion outflow model (Feng et al. 2021).In the picture of an advection-dominated accretion flow model, CL behavior might be caused by an intermittent accretion stream, which would result in a back-and-forth BEL pattern in the AGN (Noda & Done 2018).This study is first step in an ongoing effort to constrain or test these models by analyzing a large sample of CL AGN in the DESI project.

CONCLUSION
In DESI early data, we carry out a systematic search for CL AGN through cross-matching with the SDSS DR16 spectroscopic database.From a parent sample of 82,653 AGN, the following summarizes our main findings: • We have compiled a sample of 130 CL AGN selected based on VI and our defined criteria, which includes 2 Hα, 45 Hβ, 38 Mg II, 61 C III], and 10 C IV CL behaviors.In addition, we find 42 candidate CL features in confirmed CL AGN and provide another 91 CL AGN candidates, which show the dramatic flux variation of emission lines but remain a significant broad component.• The highest redshift of CL AGN in our sample is z = 3.56.• We find twenty AGN that display two simultaneous BEL changes across the two epochs, and three object that displays the simultaneous appearance of three changing BELs (Mg II, C III], and C IV). • We identify 35 CL AGN that display a turn-off transition and 95 CL AGN as a turn-on transition, whose time scales show significant differences, with turn-off larger than turn-on.
• We confirm the tendency of bluer when brighter, which is consistent with the behavior of previously discovered CL AGN.Although our current research focuses on compiling a substantial sample of CL AGN, future spectral decomposition work (such as, a more accurate black hole mass and accretion rate, a detail broad emission line measurement in the dim state) and dust extinction tests that will analyze the broad emission lines and continuum could greatly enhance our understanding of the physical mechanisms behind CL activity (Fawcett et al. 2020(Fawcett et al. , 2022).In addition, future studies of CL AGN will be crucial in advancing our understanding of the growth and evolution of supermassive black holes and the properties of the circumnuclear material in AGN (Ho 2005).The ongoing DESI project provides an exciting opportunity to gain further insight into the CL physical mechanism.In our follow-up work, we would attempt to verify the CL sequence and conduct further spectral decomposition to analyze the physical mechanism (MacLeod et al. 2016;Guo et al. 2019;Green et al. 2022).Additionally, we will continue to search for CL AGN in DESI DR1 and investigate the difference and relationship between CL AGN and normal AGN.The DESI Legacy Imaging Surveys consist of three individual and complementary projects: the Dark Energy Camera Legacy Survey (DECaLS), the Beijing-Arizona Sky Survey (BASS), and the Mayall z-band Legacy Survey (MzLS).DECaLS, BASS and MzLS together include data obtained, respectively, at the Blanco telescope, Cerro Tololo Inter-American Observatory, NSF's NOIRLab; the Bok telescope, Steward Observatory, University of Arizona; and the Mayall telescope, Kitt Peak National Observatory, NOIRLab.NOIRLab is operated by the Association of Universities for Research in Astronomy (AURA) under a cooperative agreement with the National Science Foundation.Pipeline processing and analyses of the data were supported by NOIR-Lab and the Lawrence Berkeley National Laboratory.Legacy Surveys also uses data products from the Near-Earth Object Wide-field Infrared Survey Explorer (NEOWISE), a project of the Jet Propulsion Laboratory/California Institute of Technology, funded by the National Aeronautics and Space Administration.Legacy Surveys was supported by: the Director, Office of Science, Office of High Energy Physics of the U.S. Department of Energy; the National Energy Research Scientific Computing Center, a DOE Office of Science User Facility; the U.S. National Science Foundation, Division of Astronomical Sciences; the National Astronomical Observatories of China, the Chinese Academy of Sci-ences and the Chinese National Natural Science Foundation.LBNL is managed by the Regents of the University of California under contract to the U.S. Department of Energy.The complete acknowledgments can be found at https://www.legacysurvey.org/.
Any opinions, findings, and conclusions or recommendations expressed in this material are those of the author(s) and do not necessarily reflect the views of the U. S. National Science Foundation, the U. S. Department of Energy, or any of the listed funding agencies.

Figure 1 .
Figure1.The histograms for the redshift (top) and r-band magnitude (bottom) of our parent sample (grey), selected CL AGN (blue) and CL candidates (red).The corresponding dash lines represent the median redshift values for these three samples.

Figure 2 .
Figure 2. Distribution of CL AGN in parameter spaces of Max(Nσ) vs. R (left panel) and parameter spaces of R vs. F σ,dim (right panel).The solid scatters represent the measurement of all AGN parameters (Hβ or Mg II lies in the the spectrum).The hollow triangles and pentagons represent the final CL AGN and CL Candidates after the second stage of visual inspection.Gray shading is the parameter threshods of CL AGN.The dashed gray lines are the corresponding boundary.

Figure 3 .
Figure 3. Five CL AGN example spectra of Hα (top left), Hβ (top right), Mg II (middle left), C III] (middle right) and C IV (bottom) found in DESI.Blue and red lines represent the smoothed SDSS and DESI spectra respectively (the shaded region is the corresponding original spectrum).The grey line in the lower panel is the difference between the two spectra, which is the subtraction of the dim spectrum from the bright one.The black dashed vertical lines represent the Hα, Hβ, Mg II, C III], C IV, and Lyα as if present.The red and green ticks represent the CL line feature and candidate feature respectively.
Figure 2 of Hu et al. (

Figure 4 .
Figure 4.The example spectra of CL AGN with two BEL changes during two epochs.The top left represents two BEL disappearances for both Hα and Hβ.The top right represents two BEL disappearances for both Hβ and Mg II.The bottom left represents two BEL appearances for both Mg II and C III].The top right represents two BEL vanish for both C III] and C IV.The lines in this plot are the same as in Figure 3.

Figure 5 .
photometric light curve to exclude fiber position and/or aperture issues introduced from the program selection(Yang et al. 2018;MacLeod et al. 2019;Green et al. 2022).The [O III] flux should remain relatively unchanged since the [O III] emission line probes the region reaching hundreds of parsecs to several kiloparsecs away from the central ionization source and originates from the narrow line region and host galaxy.The photometric light curve also provides a criterion to check if there is a flux calibration problem between the DESI and SDSS spectra.However, as shown in Table4, we are unable to obtain a light curve for every CL AGN and candidates, since the majority of DESI objects are fainter than the limiting magnitude of many time-domain surveys (such as ZTF, which has a limit magnitude at 20.6(Masci et al. 2019), but DESI can reach about 23 mag in the r-band).To match the resolution of the two datasets, we use a Gaussian kernel to convolve DESI spectra to the lower resolution of the SDSS spectra (R ∼ 2000).FollowingDu et al. (2018), we perform the [O III]-based calibration to scale the SDSS spectrum flux according to[O III]  flux ratio between DESI and SDSS for AGN at z ≤ 0.75.After that, we remeasure the local variation and total flux for each BEL, and remove those objects that do not satisfy the criteria outlined in Section 3.3.If the [O III] luminosity change between the DESI and SDSS spectrum keeps within the 15% deviation (e.g., due to either different fiber sizes or variational airmass), then we do not rescale the spectrum according to the [O III] calibration.For those AGN whose S/N < 3 at [O III], we determine the

Figure 6 .
Figure 6.The highest redshift z = 3.56 of C III] CL AGN found in DESI.

Figure 7 .
Figure 7. Same with Figure 3, but showing the CL candidates for Hα, Hβ, Mg II, C III], and C IV respectively.The complete figures for all CL candidates are available in their entirety in the online form.
Figure8.The histograms of the upper limit of the transition time scale for CL AGN and CL candidates in observed-frame (left panel) and rest-frame (right panel).The dash lines represent the median value of three samples.While it seems like C III] and C IV have lower rest-frame time scales than the other lines, the observed-frame histogram shows that that is partially due to a selection effect, since time-dilation has made it impossible from our sample to detect carbon rest-frame lags longer than 3500 days.
in MacLeod et al. (2019) and Figure 8 in Green et al. (

Figure 9 .
Figure 9.The histograms of the upper limit of the transition time scale for Hα + Hβ (left panel), Mg II(middle panel), C III] +C IV (right panel) CL AGN and CL candidates in rest-frame (right panel).The dash lines represent the median value of samples.

Table 2 .
The selection of CL AGN and candidates (Mg II, C III], and C IV) from DESI for z > 0.75 NOTE-The columns are the same as Table1.

Table 3 .
The information of five broad emission lines in AGN Zeltyn et al. (2022)iability between the SDSS and DESI spectra by diagnosing additional emission lines, such as Mg II, [O II], and Hα.We calibrated those objects where the [O III] luminosity deviation is larger than 15% and recalculated the selection criteria to determine whether CL or not.Finally, we reject 6 objects from the 65 Hβ CL AGN or CL candidates for the objects z ≤ 0.75.We thus find a selection accuracy of 90.8% for CL AGN.Since [O III] moves out of the optical spectrum at z > 0.75 and corresponding photometric data are not unavailable, we assume that the same accuracy applies for CL AGN sample but it could be slightly lower due to fainter spectra.Within the current DESI project, we obtain 130 CL AGN based on the variability definition outlined in Section 3.3 and VI in Section 3.2, which are listed in Table1 and 2. Of these 130 AGN, J162829.18+432948.5 was previously discovered inZeltyn et al. (2022)and J120710.25+000806.1 and 121033.30-011755.6 were selected as CL candidates reported by MacLeod et al. (2019) to have high optical variability, although were lacking spectra.Specifically, we identify 2 Hα, 45 Hβ, 38 Mg II, 61 C III]

Table 4 .
Sample properties for the CL AGN presented in this study

Table 4 .
Sample properties for the CL AGN presented in this study

Table 5 .
Sample properties for CL candidate features in confirmed CL AGN.The columns are same as Table 4 search Scientific Computing Center, a DOE Office of Science User Facility under the same contract.Additional support for DESI was provided by the U.S. National Science Foundation (NSF), Division of Astronomical Sciences under Contract No. AST-0950945 to the NSF's National Optical-Infrared Astronomy Research Laboratory; the Science and Technologies Facilities Council of the United Kingdom; the Gordon and Betty Moore Foundation; the Heising-Simons Foundation; the French Alternative Energies and Atomic Energy Commission (CEA); the National Council of Science and Technology of Mexico (CONACYT); the Ministry of Science and Innovation of Spain (MICINN), and by the DESI Member Institutions: https://www.desi.lbl.gov/collaborating-institutions.

Table 6 .
Sample properties for the CL AGN candidates presented in this study The columns are same as Table 4 (This table is available in its entirety in machine-readable form.)