Draft version August 2, 2024

Typeset using LATEX twocolumn style in AASTeX631

A closer look at dwarf galaxies exhibiting MIR variability: AGN confirmation and comparison with
non-variable dwarf galaxies
Archana Aravindan ,1 Gabriela Canalizo ,1 Nathan Secrest ,2 Shobita Satyapal ,3 and Thomas Bohn 4

1 Department of Physics and Astronomy, University of California, Riverside, 900 University Ave, Riverside CA 92521, USA
2 U.S. Naval Observatory, 3450 Massachusetts Avenue NW, Washington, DC 20392-5420, USA
arXiv:2408.00666v1 [astro-ph.GA] 1 Aug 2024

3 George Mason University, Department of Physics and Astronomy, MS3F3, 4400 University Drive, Fairfax, VA 22030, USA
4 Hiroshima Astrophysical Science Center, Hiroshima University, 1-3-1 Kagamiyama, Higashi-Hiroshima, Hiroshima 739-8526, Japan

(Received May 21, 2024; Revised July 18, 2024; Accepted July 31, 2024)

ABSTRACT
Detecting active black holes in dwarf galaxies has proven to be a challenge due to their small size
and weak electromagnetic signatures. Mid-infrared variability has emerged as a promising tool that
can be used to detect active low-mass black holes in dwarf galaxies. We analyzed 10.4 years of
photometry from the ALLW ISE/NEOW ISE multi-epoch catalogs, identifying 25 objects with AGN-
like variability. Independent confirmation of AGN activity was found in 68% of these objects using
optical and near-infrared diagnostics. Notably, we discovered a near-infrared coronal line [S IX] λ
1.252 µm in J1205, the galaxy with the lowest stellar mass (log M∗ = 7.5 M⊙ ) and low metallicity (12
+ log(O/H) = 7.46) in our sample. Additionally, we found broad Paα potentially from the BLR in
two targets, and their implied black hole masses are consistent with black hole-stellar mass relations.
Comparing non-variable galaxies with similar stellar masses and W ISE W 1 − W 2 colors, we found
no clear trends between variability and large-scale galaxy properties. However, we found that AGN
activity likely causes redder W 1 − W 2 colors in variable targets, while for the non-variable galaxies,
the contribution stems from strong star formation activity. A high incidence of optical broad lines
was also observed in variable targets. Our results suggest that mid-infrared variability is an effective
method for detecting AGN activity in low-mass galaxies and can help uncover a larger sample of active
low-mass (< 106 M⊙ ) black holes in the universe.

Keywords: Active galactic nuclei (16), Dwarf galaxies (416), Infrared galaxies (790)

1. INTRODUCTION should undergo phases of accretion at super-Eddington

The formation of the first black holes (BHs) in the rates in order to reach the masses of the SMBH that
universe is still not well understood (Volonteri & Begel- are currently present in the universe (Volonteri & Rees
man 2010; Natarajan 2014; Inayoshi et al. 2020). There 2005; Lupi et al. 2016). In an alternate scenario, the
are two most commonly discussed formation scenarios first BHs were formed by the direct collapse of gas in
for black hole seeds, which eventually evolved into su- the early universe (Silk & Rees 1998; Bromm & Loeb
permassive black holes (SMBHs) at the centers of nearly 2003; Begelman et al. 2006), and have heavier masses of
every massive galaxy. In one scenario, the first BHs were about 104 -105 M⊙ (Lodato & Natarajan 2006; Inayoshi
formed by the collapse of early Population III stars. The et al. 2020; Volonteri et al. 2021; Bogdán et al. 2024).
BHs formed by this method are likely to have masses in Other commonly suggested formation mechanisms in-
the range 100 - 1000 M⊙ (Madau & Rees 2001) and clude runaway collisions in dense star clusters (Portegies
Zwart et al. 2004; Devecchi et al. 2012; Natarajan 2021),
which can produce BH seeds with masses 103 - 105 M⊙ .
Corresponding author: Archana Aravindan Direct observational evidence for the nature of the ear-
aarav005@ucr.edu liest BHs is challenging to obtain. However, observing
2

BHs in dwarf galaxies in the nearby universe could help produces a net decrease in the predicted [O III]/Hβ
narrow down the formation scenarios (Mezcua 2017; and [N II]/Hα emission line ratios, and the positions
Reines 2022; Haidar et al. 2022; Beckmann et al. 2023). of galaxies can shift from the AGN region to the star-
If dwarf galaxies have a higher occupation fraction (per- forming (SF) region. While diagrams that use [S II] and
centage of galaxies with BHs) of low-mass (< 103 M⊙) [O I] fluxes are useful for recovering AGN across a range
BHs, it is likely that the BH seeds were formed from the of BH masses and galaxy metallicities (Cann et al. 2019;
collapse of Population III stars (van Wassenhove et al. Polimera et al. 2022), these diagrams still miss a large
2010; Greene 2012). Conversely, if they have a lower fraction of AGN (Wasleske & Baldassare 2024), most
occupation fraction of high-mass (> 105 M⊙) BHs, the likely due to the strong contribution from star formation
direct collapse scenario is favored (Miller et al. 2015; to the line ratios. The low-mass BH in dwarf galaxies
Nguyen et al. 2019; Greene et al. 2020). also have a small gravitational sphere of influence com-
Detecting active black holes, or active galactic nuclei pared to more massive BH, making them impossible to
(AGN), which are actively accreting material, is one of detect dynamically in galaxies apart from the nearby
the best methods to determine the presence of BHs in ones (Do et al. 2014; den Brok et al. 2015; Nguyen et al.
galaxies. Although AGN detections cannot provide a 2018, 2019). A low-mass BH, accreting at the Eddington
complete census of the BH occupation fraction in dwarf luminosity, will be of comparable luminosity to that pro-
galaxies (since only a small fraction of BHs are likely duced by stars in the galaxy, also making it challenging
to be actively accreting and visible as AGN), identify- to accurately detect AGNs in dwarf galaxies.
ing AGN can still offer constraints on the occupation A promising method to locate AGNs in dwarf galax-
fraction. Moreover, AGN detections can guide follow- ies is by making use of variability. AGNs are known
up spectroscopic observations to determine BH masses to be variable across the entire electromagnetic spec-
using methods such as broad-line measurements, further trum and in different time scales, ranging from months
refining our understanding of BH formation scenarios. to days depending on the mass of the galaxy (Ulrich
With advancements in telescope technology and the et al. 1997; Vanden Berk et al. 2004; Sesar et al. 2007;
advent of large-scale surveys, it has become clear that MacLeod et al. 2010; Burke et al. 2020, 2021). This
a sizeable number of dwarf galaxies are likely to host variability is often attributed to instabilities in the ac-
AGN (Reines et al. 2013; Moran et al. 2014; Chilin- cretion disk or surface temperature fluctuations (Ruan
garian et al. 2018; Mezcua et al. 2018; Mezcua & et al. 2014). Baldassare et al. (2020) made use of opti-
Domı́nguez Sánchez 2020; Salehirad et al. 2022; Mezcua cal variability using the Sloan Digital Sky Survey (SDSS,
& Domı́nguez Sánchez 2024). Results from JWST in- York et al. 2000) and Palomar Transient Factory (PTF,
dicate the presence of BH in low-metallicity galaxies in Law et al. 2009) to identify AGNs in dwarf galaxies.
the early universe (Kocevski et al. 2023; Kokorev et al. They found that nearly 75% of low-mass galaxies in their
2023; Übler et al. 2023; Harikane et al. 2023; Maiolino sample with AGN-like variability have narrow emission
et al. 2023; Goulding et al. 2023) and local dwarfs can be lines dominated by star formation and would be com-
used as laboratories in which to study similar BH, and pletely missed by the BPT diagram. Similarly Martı́nez-
their effects on the host galaxy in much greater detail Palomera et al. (2020) used the High Cadence Transit
than is possible at high z (Bose et al. 2018). Addition- Survey (HiTS, Martı́nez-Palomera et al. 2018) to look
ally, the detection of AGN in dwarf galaxies can be used for optical variability and found that nearly 95% of vari-
to constrain the BH masses and occupation fractions able AGNs in their sample do not exhibit any AGN-like
(Chadayammuri et al. 2023). optical emission line ratios. Ward et al. (2022) found 44
It is challenging to identify active BH in dwarf galaxies out of 25,714 dwarf galaxies had optically variable AGN
using traditional diagnostic methods such as the Bald- candidates using optical photometry from the Zwicky
win, Phillips & Terlevich (BPT) diagram (Baldwin et al. Transient Facility (ZTF, Bellm et al. 2019). A large sam-
1981) or Veilleux-Osterbrock (VO) diagrams (Veilleux ple of optically variable AGN in dwarf galaxies was also
& Osterbrock 1987) as dwarf galaxies are more likely to obtained from the photometric analysis of various deep
have low gas-phase metallicities (Lequeux et al. 1979). field surveys such as Dark Energy Survey (DES, Dark
Cann et al. (2019) show that, as the BH mass decreases, Energy Survey Collaboration et al. 2016; Burke et al.
the hardening of the spectral energy distribution from 2022) and COSMOS (Kimura et al. 2020), as well as
the accretion disk changes the ionization structure of predictions for possibly uncovering a larger sample from
the nebula. The enhanced high-energy emission from future surveys such as the Rubin Observatory (Ivezić
IMBHs results in a more extended partially ionized zone et al. 2019; Burke et al. 2023, 2024).
compared with models for higher-mass BH. This effect
 3

However, high levels of extinction can severely reduce do not show signs of AGN activity in the optical in or-
the effectiveness of using optical variability to uncover der to look for other AGN indicators. These include the
AGN in low-mass galaxies. Mid-infrared (MIR) vari- presence of NIR coronal lines which have proved to be a
ability, on the other hand, is less sensitive to optically useful method to identify the presence of AGN in dwarf
obscured and Compton thick AGN, which form a sig- galaxies (Bohn et al. 2021) or broad lines from the broad
nificant fraction of the low luminosity AGN population line region (BLR) that might be too obscured to de-
(Donley et al. 2008; Annuar et al. 2015, 2017; Ricci et al. tect in the optical. We also analyzed how galaxies with
2016). Ultraviolet and optical emission from the accre- variability compare to non-variable galaxies with similar
tion disk is reprocessed by the dust surrounding it and W ISE colors to study the differences, if any, between
is re-radiated in the MIR. Thus, any fluctuations from the two samples of galaxies. We do this to understand
the disk would be manifested in the MIR as well, leading the effects of any biases in host galaxy properties on the
to variability in the MIR (Kawaguchi & Mori 2011; Son observed MIR variability to indicate if MIR variability
et al. 2022; Li & Shen 2023; Son et al. 2023). is preferentially detected in certain types of galaxies.
Secrest & Satyapal (2020) found a low incidence of The paper is organized as follows: In Section 2, we
MIR variable dwarf galaxies (0.02%) using 8.4 years describe the variable sample selection, present details
of photometry from the AllW ISE/NEOW ISE multi- of the targets, and describe the observations and data
epoch catalogs and concluded that high cadence data is reduction. In Section 3, we present results of evidence
necessary for MIR selection of AGN in dwarf galaxies. of AGN activity in a subsample of MIR variable dwarfs.
Ward et al. (2022) used forward–modeled MIR photom- In Section 4, we compare the sample of variable dwarfs
etry of time-resolved W ISE coadded images to uncover to a control sample of non-variable dwarf galaxies and
148 MIR variable galaxies out of a sample of 79,879 analyze the differences between the properties of the two
dwarf galaxies. They also found that spectroscopic ap- samples. We then summarize our conclusions in Section
proaches to AGN identification would have missed 69% 5. Throughout the paper, we assume ΛCDM cosmology,
of their W ISE MIR variable dwarfs. with h = 0.73.
While AGNs are variable in the MIR, it is important
to realize that there are several other transients that are 2. DATA AND OBSERVATIONS
capable of varying in the infrared as well. Although MIR 2.1. Sample Selection
variability selection is less affected by supernovae (SN)
As in Secrest & Satyapal (2020), we created a sample
contamination than optical variability selection due to
of dwarf galaxies from the NASA-Sloan Atlas (NSA),
the fainter SEDs of SN (Smitka 2016), there has been
version nsa v1 0 1. We started by matching the full
evidence of infrared variability caused by SN and vari-
NSA catalog to AllW ISE with 10′′ , which returned a
able stars (Karambelkar et al. 2019; Sun et al. 2022;
unique match for 637168 out of 641409 (99%) of ob-
Soraisam et al. 2023). There have also been instances of
jects. We required that the elliptical Petrosian masses
infrared transients from interactions between stars and
and Sérsic masses differ by no more than 0.5 dex, which
planets (De et al. 2023), or from mergers between a pair
Secrest & Satyapal (2020) found to effectively remove
of compact objects such as neutron stars or white dwarfs
galaxies cut off in their SDSS images, leaving out 8739
(Rosswog 2005; Arcavi et al. 2017). Although analyzing
objects.
the evolution of the light curves with time can eliminate
Because we use catalog AllW ISE Multiepoch and
non-AGN sources such as rapid, one-off transient events,
NEOW ISE-R PSF photometry for the light curves,
studies have shown that intermediate-mass black holes
we require that the galaxies not be extended or con-
are capable of varying over timescales ranging from a
volved with other sources in W ISE, respectively, re-
few hours to a few days, making it difficult to distinguish
quiring ext flg<=1, nb==1. This left 459000 galaxies.
them from more rapid transients (Shin et al. 2022).
Finally, to place our sample on the W ISE W 1 − W 2,
Thus it is important to determine if MIR variability
W 2 − W 3 color-color diagram, we require that all ob-
is accurately detecting AGN, specifically in dwarf galax-
jects have valid AllW ISE photometry in the first three
ies. In this work, we aim to determine the effectiveness
bands, with corresponding cc flags equal to zero, leav-
of MIR variability in tracing AGN activity by select-
ing 316715 galaxies. Secrest & Satyapal (2020) found
ing a sample of dwarf galaxies that are variable in the
that making a cut of AllW ISE W 2 < 14.5 mag re-
MIR from Wide-field Infrared Survey Explorer (W ISE,
moves the Eddington bias from the light curves, leaving
Wright et al. 2010) light curves. We then obtained NIR
224969 systems. Of these, 6226 galaxies have elliptical
spectroscopic observations of the subset of targets that
Petrosian stellar masses less than 2 × 109 h−2 M⊙ , which
corresponds to log (M⋆ /M⊙ ) < 9.6 using h = 0.73.
 4

Table 1. Summary of AGN confirmations in variable dwarf galaxies

Target Stellar Mass AGN indicators other than MIR variability Observed with NIRES
Optical BL BPT He II Optical variability W ISE NIR Coronal lines NIR BL NIR Diagnostics
(1) (2) (3) (4) (5) (6) (7) (8) (9) (10) (11)

J000011.72+052317.4 9.29 ✓(a) ··· ··· ··· ✓ ··· ··· ··· ···
J082912.67+500652.3 9.53 ✓(b,c) ✓(e) ··· ✓(g) ✓(i) ··· ··· ··· ···
⋆
J084702.07+035202.0 9.57 ··· ··· ··· ··· ✓(h) ··· ··· ✓ ✓
J085431.18+173730.5 9.29 ✓(d) ··· ··· ··· ··· ··· ··· ··· ···
J090613.76+561015.3 9.30 ··· ✓(e) ✓ ··· ✓ ✓(j) ··· ··· ···
J092810.51+150228.0⋆ 9.47 ··· ··· ··· ··· ✓(h) ··· ✓ ✓ ✓
J093608.61+061525.9 8.29 ··· ··· ··· ··· ··· ··· ··· ··· †
J095438.79+403204.4 9.52 ✓(a, b) ··· ··· ··· ✓ ··· ··· ··· ···
J102315.19+122227.9 9.01 ··· ··· ✓ ··· ··· ··· ··· ··· †
⋆
J103206.01+225921.6 8.87 ··· ··· ··· ··· ✓(h) ··· ✓ ✓
J114433.55+072643.0 9.51 ··· ··· ··· ··· ✓ ··· ✓ ✓ ✓
J120325.67+330846.1 9.10 ··· ✓(e) ✓ ··· ✓(i) ··· ··· ··· ···
J120503.54+455151.0⋆ 7.52 ··· ··· ··· ··· ✓(h) ✓ ··· ··· ✓
J122608.78-025226.2 8.79 ··· ··· ··· ··· ··· ··· ··· ··· †
J125305.96-031258.7 8.61 ··· ··· ✓(f) ··· ··· ··· ··· ··· ···
J130819.11+434525.6 9.57 ··· ✓ ✓(f) ··· ✓ ··· ··· ··· ···
∗
J132053.67+215510.8 8.62 ··· ✓(e) ··· ··· ✓ ··· ··· ··· ✓
J135844.73+123546.9 9.34 ··· ··· ✓ ··· ··· ··· ··· ··· ✓
J140630.09-001939.3 8.93 ✓(c) ··· ··· ··· ✓ ··· ··· ··· ···
J142835.84+570834.3 9.42 ··· ··· ··· ··· ··· ··· ··· ··· ✓
J143744.58+524333.3 8.51 ··· ··· ✓(f) ··· ··· ··· ··· ··· †
J152637.36+065941.6 9.39 ··· ··· ✓(f) ··· ✓ ··· ··· ··· ···
J154404.29+275334.0 9.14 ✓ ··· ··· ✓(g) ··· ··· ··· ··· ···
J161243.20+124505.2 9.26 ··· ··· ··· ··· ··· ··· ··· ··· ✓
J233245.03-005845.8 9.45 ✓(a, b) ··· ··· ··· ✓ ··· ··· ··· ···

Table 1 continued
 Table 1 (continued)

Target Stellar Mass AGN indicators other than MIR variability Observed with NIRES
Optical BL BPT He II Optical variability W ISE NIR Coronal lines NIR BL NIR Diagnostics
(1) (2) (3) (4) (5) (6) (7) (8) (9) (10) (11)

Note—Column(1): SDSS name of the target. Column (2): Stellar Mass of the target in units of log M⊙ . Columns(3-10) AGN indica-
tors other than MIR variability. Black checkmarks indicate confirmation in literature (references indicated in parenthesis), while blue
checkmarks indicate confirmation based on the analysis carried out in this work based on the SDSS and NIRES spectrum. Column(3):
Presence of broad optical Balmer lines (a)(Veron-Cetty & Veron 2006), (b)(Greene & Ho 2007), (c)(Oh et al. 2015), (d)(Paudel et al.
2016). Column(4): Line ratios indicative of AGN activity based on the optical BPT diagram (e)(Reines et al. 2013) Column(5): Line
ratios indicative of AGN activity based on He II λ 4686 Ådiagnostic diagrams (f)(Shirazi & Brinchmann 2012) Column (6): Presence of
optical variability (g)(Baldassare et al. 2020) Column(7): W ISE colors indicative of AGN activity (h)(Sartori et al. 2015), (i)(Marleau
et al. 2017) Column(8): Presence of NIR coronal lines (j)(Bohn et al. 2021) Column(9): Presence of broad NIR Paschen lines Column(10):
Line ratios indicative of AGN activity based on NIR diagnostic diagrams. Column(11): Whether the target was observed with NIRES.
Only targets that had no other confirmation of AGN in literature (black checkmarks) were observed (see exceptions below)
⋆: Targets that only had W ISE colors indicative of AGN activity and no other AGN indicator at the time of proposing for observations.
∗ J1320 was classified as an SN candidate in spite of being classified as BPT AGN. (Izotov & Thuan 2009; Reines et al. 2013).
† Targets planned for observations but could not be observed due to poor weather.
Note: These are the reported findings from the literature for the above targets. It is possible that further targeted searches across different
wavelengths could reveal additional AGN indicators.
5
6

inspection of their light curves. We further classified the

light curves that exhibit clear variability into those with
rapid AGN-like variability and those with supernova-like
flares or slow declines in apparent magnitude.
Although we started with a sample of dwarf galax-
ies, cutting on objects with AGN-like photometric vari-
ability creates a strong selection bias on the otherwise
small fraction of contaminating background quasars and
galaxies with incorrect spectroscopic redshifts. To mit-
igate this, we uploaded the W ISE coordinates of the
variable dwarf galaxy candidates to the SDSS thumb-
nail server and removed clear instances of nearby back-
ground quasars or objects with spectra not matching
their catalog redshifts. After removing these contami-
nants, we were left with a final sample of 25 dwarf galax-
ies with bona fide, AGN-like, MIR variability (see Fig. 1
for an example). As a check, we estimated the number
of background quasars that could be contaminating our
sample of 25 objects. Using the SDSS/BOSS quasar
catalog (Lyke et al. 2020), 98% of which have W ISE
photometry (738088 objects), we find that 114119 have
Figure 1. W ISE light curves of J1205, one of the variable dwarf W 2 < 14.5 mag like our sample. As this catalog cov-
galaxies identified in this work, in different W ISE bands W1 (3.4 ers ∼ 28% of the sky, the corresponding sky density of
µm), W2 (4.6 µm) and the change in color is depicted by W1-
SDSS/BOSS quasars is ∼ 10 deg−2 . However, because
W2. The single epoch photometry of the light curve is given by
the grey points, while the blue points and the error bars represent
our AGN selection criteria is mid-IR based, we correct
the weighted mean and the standard error per 10-day observing for the obscured population that may have been largely
period. The subtle redder-when-brighter behavior, consistent with missed by the SDSS/BOSS surveys. We conservatively
the brightening being due to increasing AGN dominance over the assume that about half of the mid-IR AGN-selected
host galaxy, and the gradual nature of the light curves indicate sources are heavily obscured (e.g., Petter et al. 2023),
that the variability is likely due to AGN. giving a sky density of 20 deg−2 . The final sample of 25
objects has a maximum NSA/AllW ISE position offset
Single-epoch photometry from the AllW ISE Multi- of 4.′′ 4, of which 6023 of the 6226 dwarf galaxies satisfy.
epoch Photometry (MEP) and NEOW ISE tables for Then, the expected number of contaminant background
these 6226 dwarf galaxies were subsequently queried us- QSOs is ∼ 0.6. Thus, even if mid-IR variability selects
ing the AllW ISE source positions, using zero position all background QSOs, it is likely that there is at most
offset for MEP and a 3′′ tolerance for the NEOW ISE one contaminating background QSO in our final sample
positions. The single-epoch photometry was likewise of 25 dwarf galaxies. The final sample is presented in
cleaned, requiring valid measurements in W 1 and W 2, Table 1.
allowing a maximum deviation of ±1 mag between the The stellar masses listed in Table 1 were obtained from
single-epoch W 2 and AllW ISE W 2 corresponding to the NSA catalog (Blanton & Roweis 2007). In order
the limit set by the Eddington bias, na==0 and nb==1 for to estimate the uncertainty in these masses, we com-
uncomplicated PSF photometry, W 1 and W 2 cc flags pared the masses listed by NSA with masses reported in
and moon masked = 0, qi fact > 0 and saa sep >= 5. other catalogs. We found an average deviation of 0.01
This removed 11% of single-epoch measurements, retain- dex between the NSA masses and those reported in the
ing cleaned data for all 6226 dwarf galaxies. On average, MPA-JHU catalog (Kauffmann et al. 2003a; Salim et al.
there are 276 single-epoch photometric measurements 2007). This falls in the range of the expected deviations
for each source, from which variability metrics were cal- (-0.2 dex - 0.2 dex) between the MPA-JHU and NSA
culated as in Secrest & Satyapal (2020). In that work, stellar masses determined by Pace et al. (2019). We
we found that Pearson r between W 1 and W 2 is the also determined an average deviation of 0.3 dex between
most useful initial discriminant for genuine variability, the NSA masses and those reported by the Portsmouth
with r > 0.4 being optimal for W ISE data. This pro- group (Maraston et al. 2006) for galaxies with masses
duced an initial sample of 60 objects, allowing manual determined in all three catalogs. These values provide
 7

Table 2. Observing Log

Galaxy UT Date Obs. Redshift Stellar Mass Exp. Timea Slit PA Airmass Telluric Flux variationb
(SDSS Name) (YYYY MM DD) (log M⊙ ) (s) (degrees) (%)
J084702.07+035202.0 2023 04 07 0.059 9.57 8 × 240 306 1.05 HD 85504 15
J092810.51+150228.0 2023 04 08 0.078 9.47 4 × 240 45 1.50 HD 89239 25
J103206.01+225921.6 2023 04 08 0.058 8.87 4 × 240 0 1.50 HD 101060 50
J114433.55+072643.0 2023 04 07 0.066 9.51 4 × 240 322 1.30 HD 95126 15
J120503.54+455151.0 c 2023 04 07 0.065 7.52 6 × 240 0 1.17 HD 95126 15
J132053.67+215510.8 2023 05 08 0.022 8.62 3 × 240 30 1.28 HD 116960 <10
J135844.73+123546.9 2023 05 08 0.024 9.34 4 × 240 90 1.24 HD 122945 <10
J142835.84+570834.3 2023 05 08 0.059 9.42 6 × 240 344 1.28 HD 116405 <10
J161243.20+124505.2 2023 05 08 0.034 9.26 3 × 240 4 1.46 HD 210501 <10
a Exposures were typically done in ABBA nodding.
b Variation in flux between individual exposures.
c Indicates galaxies with coronal line detections. See Section 3.1

a measure of the average uncertainty in the masses of AGN by any method (including J1320) and four galaxies
our sample. Thus, the stellar masses that we use fall that only had W ISE colors indicative of AGN activity.
within the range of expected masses of dwarf galaxies These were the only galaxies in the sample exhibiting
even with the predicted uncertainty. Also see Fig. 2 for MIR colors indicative of AGN without any other indica-
their positions on the optical BPT diagram. tors across the criteria examined. We included the latter
We then extensively searched to see if any variable four galaxies because young starburst galaxies can often
dwarf galaxies had already been identified as AGN in mimic the mid-infrared colors of AGN galaxies (Hainline
the literature. Twelve of the 25 (48%) dwarf galax- et al. 2016) and may not truly be AGNs. Probing the
ies were already identified as AGN by a combination NIR spectra of the variable targets could uncover evi-
of other optical methods, such as the presence of broad dence of AGN activity, such as broad Paschen lines from
Balmer lines, strong He II emission, optical variability, the BLR or coronal lines indicative of a hard radiation
and emission line ratios indicative of AGN from the BPT field.
diagram (see Table 1 for references). NIR spectroscopy was obtained on three separate
Although the MIR variable galaxy J1320 has BPT line dates (UT 2023 April 7, UT 2023 April 8 and UT 2023
ratios indicative of AGN, it was also classified as a po- May 8) with the Keck II Near-Infrared Echellette Spec-
tential SN candidate in the literature (Izotov & Thuan trometer (NIRES; Wilson et al. 2004). NIRES is a
2009; Reines et al. 2013), so we do not include it among NIR echellette spectrograph with a wavelength cover-
the targets for which AGN activity was previously con- age across five bands (Z, Y, J, H, K) from 0.94 to 2.45
firmed. Additionally, only one of the 25 (J1526) vari- µm. There is a small gap in coverage between 1.85 to
able galaxies was classified as AGN from X-ray obser- 1.88 µm, but this region is dominated by water absorp-
vations (Baldassare et al. 2017). Apart from the ref- tion from the atmosphere and thus has low atmospheric
erences mentioned in the caption of Table 1, we also transmission. The average spectral resolution of the five
cross-referenced the variable targets in this work with orders ranges between 84 and 89 km s−1 (R = 3400).
Moran et al. (2014); Polimera et al. (2022) for BPT Observations on UT 2023 April 7 and UT 2023 April 8
AGN, Chilingarian et al. (2018) for optical broad lines, were taken under variable and heavy cloud cover, while
and Bi et al. (2020); Ohlson et al. (2024) for X-ray data, observations on UT 2023 May 8 were taken through light
but we did not find any matches. cirrus. Due to unfavorable weather conditions, we could
get data for only nine of the 13 targets. Individual expo-
2.2. Observations and Reductions sures for each target were taken for 4 minutes using the
We selected a subsample of variable dwarf galaxies for standard ABBA nodding, amounting to total exposure
observations in the NIR. The 13 galaxies in the sub- times ranging from 12 to 32 minutes, depending on the
sample include nine galaxies with no confirmations of brightness of the target. A0 spectral class stars were
8

1.5
Kewley et al. 2001 Kewley et al. 2006 Kewley et al. 2006
Kauffmann et al. 2003 Observed with NIRES
1.0 Stasinka et al. 2006 Seyferts Not observed with NIRES
log([OIII]/Hβ)

0.5

SF AGN

0.0
SF LINERS

−0.5

−1.0
−2.0 −1.5 −1.0 −0.5 0.0 0.5−2.0 −1.5 −1.0 −0.5 0.0 0.5 −2.5 −2.0 −1.5 −1.0 −0.5 0.0
log([NII]/Hα) log([SII]/Hα) log([OI]/Hα)

Figure 2. Optical BPT and VO diagrams showing the positions of the variable targets observed with NIRES (blue) and the variable
galaxies not observed with NIRES (cyan). The emission line ratios were derived from fluxes obtained from the SDSS MPA-JHU catalog.
Most of the observed variable targets lie primarily in the SF regions, and there is no indicator of AGN activity based on these diagnostic
diagrams.

used as telluric standard stars, with measured magni- are listed in Table 2). This introduced uncertainty in
tudes in J, H, and K bands. Observations of the telluric the flux calibration, as the latter was done relative to
stars were taken before or after each science target to telluric standards observed immediately before or after
correct for the atmospheric absorption features. A sum- each science target. We estimated that the uncertainties
mary of the observations is shown in Table 2. are of the order of the flux variations, and we included
The data were reduced using two modified pipelines. these added uncertainties in the measurements of fluxes.
The flat fielding and background subtraction were done
by the first pipeline using techniques described in Kelson 2.3. Spectral fitting
(2003) and Becker et al. (2009). Rectification, telluric All NIR spectra were fit using the Bayesian AGN
correction, wavelength calibration, and extraction were Decomposition Analysis for SDSS Spectra (BADASS,
done with a slightly modified version of REDSPEC 1 . Sexton et al. 2021) code, which was modified to cover
The telluric correction was done by dividing each sci- the NIR wavelength regime. A narrow component for
ence target spectrum by the spectrum of its correspond- each emission line was fit simultaneously with a second-
ing telluric star and multiplying it by a blackbody curve order polynomial for the continuum. Absorption fea-
of the same temperature. Wavelength calibration was tures were fit using the penalized pixel fitting method
done using strong OH lines, and the 1D spectra were (pPXF, Cappellari & Emsellem (2004)) with templates
then median combined. Flux calibration of individual from the eMILES (Vazdekis et al. 2016) stellar library.
targets was done by converting the magnitude of the The error (noise) in the fit was determined by calcu-
telluric star to the associated flux in that band 2 . A lating the standard deviation of the continuum flux, rep-
minor corrective factor (<5%) was introduced due to resented by the grey-shaded region in the lower panel of
the differences between the center of the NIR bands and Figure 3 (b). The σresiduals , also indicated in the residual
the center of the wavelength coverage. Because we ob- plot (lower panel) of Figure 3(b) denotes the standard
served through clouds, we found 15 - 50 % variations deviation of the residuals obtained by subtracting the
in the flux levels between individual exposures for the modeled spectrum from the observed data.
first two nights and less than 10 % variation in flux on
the final night (Flux variations for individual objects 3. EVIDENCE OF AGN ACTIVITY
3.1. Detection of coronal lines in J1205
1 https://www2.keck.hawaii.edu/inst/nirspec/redspec.html Coronal lines (CL) are high ionization fine-structure
2 https://irsa.ipac.caltech.edu/data/SPITZER/docs/ lines that arise from collisionally excited forbidden fine-
dataanalysistools/tools/pet/magtojy/ structure transitions in highly ionized species, with ion-
ization potentials that extend well beyond the Lyman
 9
fλ (10−17 erg cm−2 s−1 Å−1 )
8

1.20 1.22 1.24 1.26 1.28 1.30 1.32

Rest Wavelength (µm)

6 Narrow/Core Comp.
[S IX] [Fe II]
fλ (10−17 erg cm−2 s−1 Å )

Host/Stellar
−1

Data
4
Model

−2
2.5 σresid = 0.6598
∆fλ

0.0
−2.5
1.240 1.245 1.250 1.255 1.260
Rest Wavelength, λ (µm)

Figure 3. (a) NIR spectrum of J1205 in order 5 of NIRES, shifted to rest-frame wavelength. The blue-dashed line indicates the coronal
line [S IX] detected at a 5σ level. The grey-shaded area represents the region of significant telluric absorption. (b) The zoomed-in spectrum
with spectral fits to the [S IX] (1.252 µm) line and the [Fe II] (1.257 µm)line. The lower panel depicts the residuals to the fit, with the 1σ
error in the spectrum shaded in grey around the dashed line indicating 0. The σresiduals indicates the standard deviation of the residuals
determined by subtracting the fit to the spectrum from the data.

limit to several hundred electron volts (eV). They are by the standard deviation of the posterior distribution
typically attributed to AGN activity and have been suc- from the MCMC fitting of the lines. Based on the com-
cessfully detected in dwarf galaxies hosting AGN (e.g., monly observed NIR emission lines in galaxies (Riffel
Bohn et al. 2021) We looked for NIR coronal lines in the et al. 2006; Landt et al. 2008; Martins et al. 2013), no
nine galaxies observed with NIRES. other emission lines are likely to be found at that wave-
The NIR coronal line [S IX] λ1.252 µm (I.P. 328.2 eV length.
and critical density log (ne )=9.4 cm−3 ) was detected in The detection of [S IX] in J1205 is tantalizing evidence
J1205. The fit to the spectrum showing the [S IX] is that it likely hosts an AGN. This is the first instance of
shown in Fig. 3. The [S IX] line is a 5σ detection, and a NIR coronal line being discovered in J1205, provid-
the measured flux of the line is 2.11 ± 0.52 × 10−16 erg ing strong evidence that J1205 has an active black hole.
cm−2 s−1 . The full width at half maximum (FWHM) J1205 has the lowest stellar mass and metallicity among
of the line is 138 ± 10 km s−1 , which is greater than all the targets (log(M∗ ) = 7.5 M⊙ and 12 + log(O/H) =
the spectral resolution of NIRES in that order (85 km 7.46; Izotov et al. 2021). It was classified as a compact
s−1 ) and thus is unlikely to be a noise spike or a cosmic starforming galaxy with extremely high [O III]/[O II]
ray signal. The errors in the flux and FWHM are given flux ratios by Izotov et al. (2017) and shown to have a
10

[Ne V] λ3426 Å detection and evidence of hard ioniza- s−1 , which is below the 3σ limit (see Fig. 5). Thus, this
tion by Izotov et al. (2021). Spectroscopic fits to the Hα feature does not appear to be a 3σ detection, but it is
emission line in spectra obtained from the Large Binoc- possible that with deeper observations, this feature can
ular Telescope (LBT) indicate the presence of a broad be detected with greater confidence.
line feature, which is attributed to expansion motions of While the detection of the coronal line [S IX] alone
supernova remnants (Izotov et al. 2017). Harish et al. and the absence of [Si VI], the most commonly observed
(2023) classified J1205 as a green pea galaxy possibly coronal line in galaxies (Rodrı́guez-Ardila et al. 2011;
hosting an AGN based on MIR colors and variability Lamperti et al. 2017; Müller-Sánchez et al. 2018; Bohn
from W ISE. The low metallicity and stellar mass of et al. 2021), may seem surprising, such occurrences are
J1205 would have made it impossible to detect the pres- not unprecedented. There have been a few instances of
ence of the AGN by traditional methods such as the galaxies that have a [S IX] detection, but only upper lim-
BPT diagram. This indicates that MIR variability can its to the [Si VI] flux are reported (Lamperti et al. 2017;
be a promising tool and, sometimes, the only method to den Brok et al. 2022). Comparing the [Si VI] to [S IX]
detect AGN in low-mass galaxies. flux ratios recorded in the literature (Rodrı́guez-Ardila
et al. 2011; Lamperti et al. 2017; Bohn et al. 2021; den
fλ (10−17 erg cm−2 s−1 Å−1 )

Brok et al. 2022), which range from 0.1 to 8 (more than

8
half the galaxies have ratios less than 3.5), our measured
H2
ratio of 0.25 (assuming the 3σ flux limit for [Si VI]) falls
6 [Si VI] ?
within this observed range. It is possible that we simply
did not obtain sufficiently deep observations to detect
4 the [Si VI] line in J1205.
No coronal lines were detected in the remaining eight
2 targets observed with NIRES, potentially due to the rel-
atively low S/N of the data. To investigate this, we plot-
0 ted the upper limits to the [Si VI] flux for each target
1.950 1.955 1.960 1.965 1.970
versus the W ISE W2 (4.6 µm) flux in Fig. 5. We de-
Rest Wavelength (µm) termined the upper limits to the [Si VI] fluxes in the
non-detections by estimating the 1σ noise levels in the
Figure 4. Keck/NIRES spectrum of J1205 in the wavelength region around [Si VI] flux and multiplying it by three to
region around [Si VI]. The small emission feature found at the obtain the 3σ upper limit. The solid line in Fig. 5in-
wavelength of [Si VI] (indicated by the dashed blue line) has a
dicates the best-fit relation, which was derived based
flux compatible with the expected flux for [Si VI], however, it is
below the 3σ detection limit and may not be real.
on observations from Müller-Sánchez et al. (2018) (in-
dicated as black dots) by Cann et al. (2021). The plot
indicates that J1428 is the only target for which [Si VI]
While the presence of [S IX] in J1205 is likely to indi-
emission is highly unlikely to be present, as its upper
cate AGN activity, it could also possibly be the result of
limits are comfortably below the scatter of the relation,
a tidal disruption event (TDE; Hinkle et al. 2023; Clark
and it would have been detected if it were present. For
et al. 2024). This can be determined by follow-up obser-
the rest of the targets, we would likely require deeper
vations to check if the coronal line persists or declines
observations in order to detect [Si VI]. The absence of
after an interval of time. However, a TDE would still
CLs in these targets is unlikely to be due to high ex-
indicate the presence of a black hole in J1205.
tinction, as J1205 (the target with CL detection) does
We did not detect any other coronal lines apart from
not have significantly lower extinction (calculated from
[S IX] in J1205. It is possible that circumnuclear stel-
the Hα over Hβ Balmer decrement) than the rest of the
lar populations dominate the NIR continuum and thus
observed targets.
drown out the CL emission, making fainter lines more
While the presence of CLs in a galaxy’s spectrum con-
difficult to detect (Rodrı́guez-Ardila et al. 2011), espe-
firms the existence of a black hole, the absence of CLs
cially since J1205 has been classified as a compact star-
does not mean the contrary. Bohn et al. (2021) analyzed
forming galaxy with very young ages (< 3 Myr) for star
a sample of 9 dwarf galaxies with optical and NIR line
formation bursts. There is a feature near the noise limit
ratios indicative of AGN and found NIR CLs only in 5 of
that is suggestive of [Si VI] (see Fig. 4), but the feature
them (55 %). Several other works in the literature have
did not have sufficient signal to be fit using BADASS.
found less than 100% CL detections in samples of Sy1
We independently fit a Gaussian to this feature and de-
and Sy2 galaxies with similar spectral windows: 66%
termine the flux to be 1.78 ± 0.88 × 10−17 erg cm−2
 11

Table 3. Black hole masses

−13
MS+18
Cann+2021 LPaα FWHM log10 (MBH ) log10 (Mstellar )
CL([S IX]) detection (erg s−1 ) (km s−1 ) (M⊙ ) (M⊙ )
−14
CL non-detections J0928 1042.5±0.2 325 ± 5 6.24 ± 0.27 9.47
log([Si VI] 1.963 µm)

J1144 1041.2±0.12 457 ± 5 5.96 ± 0.14 9.51

−15 3 2
order) model in order to justify the number of Gaussian
7 components we include in our fit. To do this, we cal-
4
culated the standard deviation of the residuals of the
−16 1 fit (given by σ). We performed the F- test, given as
9 5 F = (σlower−order )2 /(σhigher−order )2 , for one versus two
6
components for the Gaussian fits to the Paα line. The F-
−17 test was calculated over a wavelength range of 1.86–1.89
8 µm, which includes the Paα λ 1.876 µm. An F value
greater than 3 indicates that the higher-order fit (i.e.,
the fit using more components) is justifiable. Using this
−18
−14 −13 −12 −11 −10 method, we determined that the Paα emission lines in
log(W2 4.6 µm)) J0928 and J1144 both required a second Gaussian com-
ponent, which we allowed to freely vary to fit the broad
feature present in the emission lines using BADASS. The
fits to the spectrum can be seen in Fig. 6.
Figure 5. Plot of [Si VI] vs. W2 flux for the [Si VI] non-
detections in our sample. The solid line indicating the correlation Both J0928 and J1144 show a broad component to
is from Cann et al. (2021), which was derived from observations Paα while not showing a broad component in the opti-
from Müller-Sánchez et al. (2018) and the shaded region indicates cal Balmer lines (at least at the spatial resolution of the
the scatter in the data. The targets are marked as J0847 (1), SDSS spectrum, observed with a fiber that includes sig-
J0928 (2), J1032 (3), J1144 (4), J1205 (5), J1320 (6), J1358 (7), nificant contaminating/diluting stellar light). In galax-
J1428 (8), J1612 (9). Targets whose upper limits are above the ies with strong dust obscuration, it is possible that the
relation would require deeper observations in order to detect the
broad line region (BLR) does not manifest itself in op-
[Si VI] line, while for the targets with upper limits below the re-
lation, it is likely that the [Si VI] line is not present. For more
tical emission lines. However, NIR wavelengths are less
details regarding the upper limits, refer to the appendix affected by dust extinction, making it easier to observe
broad components that would be obscured at optical
(36 out of 54, Rodrı́guez-Ardila et al. (2011)), 25%, (5 wavelengths. This method has been used to uncover op-
out of 20, Mason et al. (2015)), and 43% (44 out of 102, tically hidden AGN in massive galaxies (Veilleux et al.
Lamperti et al. (2017)). 1997; Lamperti et al. 2017; Bohn et al. 2020). There
The relatively low detection rates in all of these sam- were no indicators of a broad component being present
ples could be an observational effect, as ground-based in any of the forbidden lines in the NIR for the two tar-
observations are always hampered by telluric absorption gets. Combined with the fact that the SDSS spectrum
from the atmosphere. Depending on a target’s redshift, of the two targets also did not have a significant broad
the CLs can be shifted into regions of strong telluric component in the forbidden [O III] lines, it is unlikely
absorption and/or in regions dominated by strong OH that the broad component is produced by an outflow.
skylines, making it difficult to detect weak CLs. Other Assuming the broad lines measured are from the BLR,
factors include a dominant contribution from the stellar we used the FWHM and luminosity of the broad Paα
population to the NIR continuum and the presence of component to estimate black hole masses. Using virial
stellar absorption features. mass estimators from Kim et al. (2018), we determined
3.2. Detection of broad Paα lines the black hole mass (Table 3) from the following equa-
tion:
Two of the variable targets (J0928 and J1144) show
a broad line component in Paα at 1.876 µm. We jus-
0.49 2
tify the need for a second component based on the F-
 
M LPaα FWHM
= 107.07 ×
test. The F-test determines the significance of the fit M⊙ 1042 erg s−1 103 km s−1
between a complex (higher-order) and a simple (lower- (1)
12

Narrow/Core Comp.
2000 Broad Comp.
fλ (10−17 erg cm−2 s−1 Å )
−1

Host/Stellar
Data
1500
Model

1000

500
He I
0
50
σresid = 10.9590
∆fλ

−50
1.860 1.865 1.870 1.875 1.880 1.885 1.890
Rest Wavelength, λ (µm)

(a)

Paα
Narrow/Core Comp.
Broad Comp.
fλ (10−17 erg cm−2 s−1 Å )

300
−1

Host/Stellar
Data
200 Model

100
He I

0
5 σresid = 1.8731
∆fλ

0
−5
1.860 1.865 1.870 1.875 1.880 1.885
Rest Wavelength, λ (µm)

(b)

Figure 6. Broad-line fits to the Paα line in two variable targets, J0928 (a) and J1144 (b), which could indicate a broad component from
the BLR of the galaxies. There was an unidentified line next to He I line at 1.869 µm, which was fit but left unmarked. The lower panels
in both figures depict the residuals to the fit, with the 1σ error from the noise in the spectrum shaded in grey around the dashed line
indicating 0.

The errors given in Table 3 are the standard devia- guish between AGN and SF regions in galaxies. How-
tion of the posterior distribution from the MCMC fit- ever, similar diagnostic diagrams in the NIR are scarce,
ting of the lines. The measured black hole masses agree chiefly due to a lack of large spectroscopic surveys in the
with the predicted black hole masses based on the stel- infrared. Nevertheless, there have been several attempts
lar masses of J1144 and J0928 (Fig. 7) from Reines & to recreate similar diagnostic diagrams in the infrared
Volonteri (2015). that can be used to distinguish between AGN and SF
regions (Larkin et al. 1998; Smith et al. 2014; Colina
3.3. NIR diagnostic diagrams et al. 2015). In particular, Calabrò et al. (2023) used
Optical diagnostic diagrams like the BPT diagram CLOUDY photoionization models to identify AGN - SF
have been well studied and used extensively to distin- diagnostics based on the ratio of bright near-infrared
 13

Most of the MIR variable targets in our sample fall

11 in the SF region of the optical BPT diagram as seen in
Fig. 2. We plotted them in the various NIR diagnos-
10 tic diagrams in Fig. 8. We primarily used the ratios of
[S III] λ9530 Å/Paγ and [S III] λ9530 Å/Paβ to [Fe II]
9 λ16470 Å/Paα and [Fe II] λ12570 Å/Paβ, obtained from
log(MBH /M )

NIRES. For four of the targets (marked as filled circles

in Fig. 8), these were the lines with signal-to-noise ratios
8 (S/N) > 3 in our sample, and their flux values could be
estimated with minimum uncertainties. For the remain-
7 ing targets, we could only determine one of the ratios as
at least one of the emission lines on the x-axis ([Fe II]
6 1.64µm, [Fe II] 1.25µm, Paα or Paβ) fell in the telluric
region or were in a noisy part of the spectrum, and thus
RV15 we could not measure the fluxes. We instead find the
5 upper limits (if the [Fe II] lines were undetermined) or
This work lower limits (if the Paschen lines were undetermined)
4 to the emission line ratios and showed the targets as
9 10 11 12 open circles with arrows to indicate upper or lower lim-
log(M∗ /M ) its (both if neither of the line fluxes used in the ratio
could be determined) to the ratios in Fig. 8. The upper
limits for J1032, J1428, and J1320 did not exclude the
possibility that these targets are AGN.
Figure 7. Black hole vs. stellar mass for a sample of galaxies
at low-redshift. The yellow points are galaxies at a redshift < We found that four of the variable targets (J0847,
0.06 and include dwarf galaxies with broad lines from Reines & J0928, J1144, and J1358) fell in the composite and AGN
Volonteri (2015) (RV15). The solid line is the expected linear regions across all four NIR diagnostic diagrams. J0928,
regression from Equations 4 and 5 in RV15, along with the 0.5 dex which has a broad Paα line, moved from the SF region
error, indicated by the gray shaded region. The two blue points in the optical to the AGN region of two of the NIR di-
indicate the calculated BH masses based on the broad Paα for agnostic diagrams. The other broad line target, J1144,
J0928 and J1144, respectively, which fits well with the expected
also moved into the composite region of all four NIR
black hole mass for the two galaxies.
diagnostic diagrams.
While NIR diagnostic diagrams appear to be more
emission lines such as [S III] 9530 Å, [C I] 9850 Å, [P II] effective in some cases at identifying obscured AGN in
11880 Å, [Fe II] 12570 Å, and [Fe II] 16470 Å to the dwarf galaxies, they are still limited by the gas-phase
Paschen lines. They used a sample of 130 AGN and SF metallicities of the galaxies. J1205, with a coronal line
galaxies with redshifts 0 < z < 3 observed using JWST- indicating the presence of a black hole, still lies firmly
NIRSpec (Jakobsen et al. 2022) as part of the Cosmic in the SF region of all the NIR diagnostic diagrams.
Evolution Early Release Science Survey (CEERS; PID
1345, PI Finkelstein, Finkelstein et al. 2023) and found
that this diagram is able to successfully distinguish be- 3.4. Ruling out supernova contribution to variability
tween AGN and SF galaxies. Furthermore, they found Although the nine galaxies that were observed using
60% more AGN using NIR diagnostic diagrams than NIRES are determined to exhibit AGN-like variability
when using optical BPT diagrams. Thus NIR diag- from the light curves, our selection method does not
nostic diagrams are able to uncover optically ‘hidden’ completely rule out contributions from SN. Five of the
AGNs, thus highlighting the importance of using NIR galaxies have alternate evidence of AGN activity based
wavelengths to potentially locate AGN in dusty galax- on the presence of coronal lines, broad Paα lines, and
ies. Since optical emission lines are significantly affected emission line ratios indicative of AGN, which makes it
by extinction, it is possible to overlook AGN indicators unlikely that there is a significant contribution from SN
from optical emission lines due to extinction. On the to the variability. However, Baldassare et al. (2016)
other hand, NIR emission lines are less sensitive to dust showed that in the optical, broad Hα lines could be
and thus could prove to be more helpful in identifying caused by SN activity. The ideal method to rule out the
AGN activity in dusty galaxies. possibility of broad lines being caused by SN is to ob-
tain multi-epoch spectroscopy that would show whether
14

Variable dwarfs (This work) Type 2 AGN (Riffel+06) Type 2 AGN (Riffel+19) SF max limit (Calabro+23)
Dwarf AGN (Bohn+21) Starburst (Riffel+06) Starburst (Riffel+19) AGN min limit(Calabro+23)
2.0 2.0

1.5
1.5
Composite
4
1.0 7 4
log([SIII]9530 Å/Paγ)

2 1 1.0 7
3 AGN 1 2
6 9
0.5 3
9 6
8 0.5

SF
8
0.0
5 Z
0.0
log(U) 5
−0.5

−0.5
−3.0 −2.5 −2.0 −1.5 −1.0 −0.5 0.0 −3.0 −2.5 −2.0 −1.5 −1.0 −0.5 0.0 0.5

1.5 1.5
log([SIII]9530 Å/Paβ)

1.0 1.0

7 7
2 2
0.5
6 31 0.5
3 61
94 9 4

0.0 8 0.0 8

5 5
−0.5 −0.5
−3.0 −2.5 −2.0 −1.5 −1.0 −0.5 0.0 −2.5 −2.0 −1.5 −1.0 −0.5 0.0 0.5
log([FeII]1.64 µm/Paα) log([FeII]1.257 µm/Paβ)

Figure 8. NIR diagnostic diagrams using AGN and SF limits from Calabrò et al. (2023). The targets are marked as J0847 (1), J0928 (2),
J1032 (3), J1144 (4), J1205 (5), J1320 (6), J1358 (7), J1428 (8), J1612 (9). For the targets where some of the line fluxes were not available,
we plotted them as open circles with arrows, indicating upper and lower limits. Dwarf Galaxies with AGN from Bohn et al. (2021) and
Type 2 AGN from Riffel et al. (2006) are plotted in red and green respectively. The gas-phase metallicity (Z) increases from bottom to
top, and the ionization parameter (log (U)) increases from right to left. Four of the targets moved to the composite/AGN region across all
four NIR diagnostic diagrams, as compared to their positions in the SF region in the optical diagnostic diagrams.

the broad emission lines persist. Although we do not for the common NIR emission lines associated with SN
have multi-epoch spectra and cannot check for line vari- (Rho et al. 2018) and did not find any features that
ability, we did inspect all the light curves by eye, and indicated the presence of SN.
determined that they do not show the rapid increase
and decline in their curves that is characteristic of su-
pernovae. Furthermore, we inspected the NIR spectrum 3.5. Summary of results
 15

25 MIR Variable
AGN candidates

12 classified as
9 observed in the
optical AGN in 4 not observed
NIR
literature

2 with broad lines and 2 lie in the

lie in the composite/AGN
1 with coronal line composite/AGN region
4 Inconclusive
region using NIR
using NIR diagnostics diagnostics

Figure 9. Flowchart summarizing the results characterizing the effectiveness of MIR variability in tracing AGN activity in low mass
galaxies.

Dwarf galaxies with active black holes have been found activity. The low luminosity of the AGN can lead to the
to not display AGN signatures uniformly across differ- possibility that SF activity capable of producing similar
ent detection methods. If we have a dwarf galaxy that luminosities often act as interlopers. In the infrared in
is variable in the MIR, how reliably can we determine particular, Hainline et al. (2016) found that most of the
that it is actually hosting an active black hole? In our BPT-confirmed AGNs from Reines et al. (2013) were not
work, we found that out of 25 dwarf galaxies that exhibit selected as W ISE AGN, which could be due to the host
MIR variability, we have independent evidence that 17 galaxy light dominating in the mid-infrared color space.
of them (68%) host an AGN, strongly suggesting that Conversely, star-forming dwarf galaxies are capable of
MIR variability is due to the presence of an AGN in heating dust in such a way that mimics the infrared
these targets. Twelve of these were classified as opti- colors of more luminous AGNs. That is why W 1 − W 2
cal AGN, and the remaining five had NIR signatures of colors alone should not be used to select for AGN in
hosting an AGN. Four did not have any optical indica- dwarf galaxies.
tors of AGN, but we were not able to obtain NIR obser-
vations for them, so the true fraction of independently 4. COMPARISON WITH NON-VARIABLE DWARF
confirmed AGN in the sample of MIR variable AGN can GALAXIES
be higher than 68% (see Fig. 9 for a summary).
Given the potential contamination from SF processes
Our results showed that if a galaxy is variable in
at infrared wavelengths, it is crucial to test whether MIR
the MIR, there is a high probability that it is actu-
variability is accurately tracing AGN activity or if it is
ally hosting an active black hole. Using TESS light
being caused by SF interlopers in dwarf galaxies. We
curves, Treiber et al. (2023) identified a similar fraction
have found that MIR variability is frequently accompa-
(62%) of optically variable galaxies that were either pre-
nied by alternate indicators of AGN (see Sec. 2.1 and
viously classified as AGNs in the literature or confirmed
Sec. 4), making it less likely that it is caused by stel-
as AGNs based on a combination of emission-line di-
lar processes. Additionally, we compared dwarf galaxies
agnostics, mid-IR colors, or X-ray luminosity in their
that exhibit MIR variability with similar dwarf galax-
analysis. Thus, variability is a good indicator of AGN
ies that do not exhibit MIR variability. We did this to
activity in dwarf galaxies and can be used to recover a
recover differences in optical and infrared galaxy prop-
larger fraction of AGN that is missed by other methods.
erties, if any, between the two samples of galaxies and
Signs of active black holes in dwarf galaxies are often
understand the effects of any potential biases caused by
difficult to distinguish from those originating from SF
host galaxy properties on the observed MIR variability.
16

40 40
Variable targets
35 35
Non-variable controls
30 30

25 25

20 20

15 15

10 10

5 5

0.0 0.5 1.0 1.5 2.0 7.5 8.0 8.5 9.0 9.5
(W1-W2) log M∗

Figure 10. Histograms showing the distribution of the W ISE W 1 − W 2 colors and stellar masses of the target sample of variable dwarf
galaxies and the control sample of non-variable dwarf galaxies.

4.1. Control sample selection vidual galaxies from the Hα/Hβ Balmer decrement. We
We selected a control sample of non-variable dwarf then obtained extinction corrected [O III] λ5007 Å lu-
galaxies, which were matched in W ISE W 1 − W 2 col- minosity values for all the galaxies in the target and the
ors and stellar mass with the sample of variable dwarf control samples. We also measured the star-formation
galaxies. We obtained five control galaxies on average rates (SFRs) for the galaxies using a number of meth-
for each of the 25 target galaxies that we initially classi- ods. We directly obtained the SFRs for the galaxies from
fied as variable and thus created a control sample of 136 the MPA-JHU catalog, which were calculated based on
non-variable dwarf galaxies. A two-sample Kolmogorov- the methods used in Brinchmann et al. (2004). We inde-
Smirnov (K-S) test between the W 1 − W 2 colors and pendently estimated SFRs from the extinction-corrected
stellar masses of the variable galaxies and non-variable Hα luminosity values based on the relations by Ken-
control galaxies gives p=0.99 and p=0.56, respectively, nicutt (1998) as well as from the extinction-corrected
indicating that the control galaxies do not have system- [O II] fluxes using Equation 11 in Kewley et al. (2004).
atically different W ISE colors or stellar masses (see The SFRs calculated by these different optical methods
Fig. 10). None of the galaxies in the control sample agreed with each other. We determined the optical col-
passed the variability tests based on our initial con- ors (u-r) from SDSS for the different galaxies.
straints (see Section 2.1), and thus can be safely assumed
to be non-variable. The two samples have similar red-
shift distributions, ranging from 0.01 to 0.1. 4.2. Large-scale properties of variable and non-variable
We then obtained emission line flux values for [O II] dwarf galaxies
λλ3727,3729 Å, He II λ4686, Hβ λ4861 Å, [O III] λ Comparing various properties of variable and non-
5007 Å, [O I] λ 6300 Å, [N II] λ6584 Å, Hα λ6563 Å and variable targets with similar W 1 − W 2 colors, we did
[S II] λ6731 Å from the MPA-JHU catalog (Kauffmann not find a significant difference between the extinction
et al. 2003a). We measured extinction values for indi- values, [O III] luminosities, optical colors, and SFRs be-
tween both samples (Fig. 11).
 17

35
Variable
30
60 Non-variable
25

40 20

15

20 10

0
0.0 0.5 1.0 1.5 4 6 8
E(B-V) log10 (L[OIII] /L )

35 35

30 30

25 25

20 20

15 15

10 10

5 5

0 1 2 3 −2 −1 0 1
(u-r) SFR (MPA − JHU) [log10 (M /yr)]

Figure 11. Histograms showing the comparison optical properties of the variable and the non-variable dwarf galaxies. The dashed lines
indicate the average values for the two samples. We did not find a significant difference in several key optical properties, such as extinction,
[O III] luminosity, optical colors and SFRs.
18

1.5
Kewley et al. 2001 Variable
Kauffmann et al. 2003 Non-variable
Variable 400
Non-variable
1.0

[OIII] FWHM (km s−1 )

300
log([OIII]/Hβ)

0.5

AGN 200

0.0
SF

100

−0.5

−1.0
−2.0 −1.5 −1.0 −0.5 0.0 0.5
0 250 500 750 1000 1250 1500
log([NII]/Hα) −1
Hα FWHM (km s )

Figure 12. The BPT diagram of the variable and non-variable

Figure 13. Plot of the Hα FWHM vs the [O III] FWHM for the
samples. Galaxies with log [N II]/Hα < -1 (indicative of low
variable and non-variable targets. A larger fraction of variable
metallicity) are in the blue shaded region). Using log [N II]/Hα as
targets have very broad (FWHM > 500 km s−1 ) Balmer lines
a proxy for metallicity, we found that a larger fraction of variable
than non-variable targets.
dwarfs have higher metallicities than non-variable dwarfs.

Using the emission line values obtained from the MPA- Hβ (R23, Pagel et al. 1979), particularly in the range of
JHU catalog, we plotted the BPT diagrams of both sam- metallicities (7.5 < 12 + log(O/H) < 8.5) that we would
ples (see Fig. 12). We found that 3/25 (12%) of the vari- expect these dwarf galaxies to have (Maiolino et al. 2008;
able targets lie above the Kauffmann et al. (2003b) de- Jones et al. 2015). Thus, to comment on the metallicities
marcation line for the AGN region, while 8/136 (0.05%) of the targets, we looked solely at the [N II]/Hα line
of the non-variable controls lie in the AGN region. ratios from the BPT diagram, which can be used as a
We obtained He II λ4686 emission line fluxes for proxy for metallicity (Kewley & Dopita 2002; Pettini &
eight of the variable galaxies and eighty-two of the non- Pagel 2004; Kewley & Ellison 2008).
variable galaxies from the MPA-JHU catalog. No signif- We found that, on average, variable targets have
icant difference was found between the positions of the higher metallicities (log [N II]/Hα > -1) as compared
variable and non-variable galaxies in the He IIλ4686/Hβ to the non-variable targets. Only 2/25 (8%) targets
versus [N II]λ6584/Hα diagnostic diagram presented in have [N II]/Hα lower than -1 as compared to 59/136
Shirazi & Brinchmann (2012). While some of the vari- (43%) for the control sample, as seen in Fig. 12. BPT-
able galaxies were previously identified as AGNs using selected AGNs in dwarf galaxies are expected to be bi-
this diagnostic, we identified four additional variable ased towards higher metallicities (Reines et al. 2013).
galaxies that meet the AGN criteria. These galaxies However, we adopted a completely independent method
are marked with blue checkmarks in Column 5 in Table of selecting AGN in dwarf galaxies, and we still found
1. that variable dwarf galaxies (tracing the AGN) have
Determining gas-phase metallicities for the targets is higher metallicities than non-variable dwarfs. Since
challenging as the traditional methods that depend on AGN contribution can also lead to an increase in the
various emission lines such as [O III] and [O II] often do [N II]/Hα ratio, a large fraction of variable targets with
not account for the contamination due to AGN activity. high [N II]/Hα ratios may be a direct consequence of the
Moreover, studies have shown that there is an ambiguity presence of AGNs in the variable targets.
in determining metallicities using methods such as the
ratio of [O II]λλ 3726,3729 + [O III]λλ4959,5007 over 4.3. Optical broad lines and W 2 − W 3 colors
 19

We obtained the SDSS spectra of all the galaxies in

the target and the control samples. We then performed 20

spectral fits for the entire sample in the regions that in- 0
1.5 2.0 2.5 3.0 3.5 4.0 4.5 5.0
clude the prominent Balmer lines (4400 - 7000 Å) and fit
simultaneously for Hβ λ 4861 Å, [O III] λλ 4963, 5007 Å, 2.00 2.00

Hα λ 4861 Å, and [N II] λλ 6548, 6583 Å. We found that 1.75
AGN Demarcation Box
1.75

variable targets have broader Balmer lines compared to 1.50 1.50

non-variable targets (Fig. 13): 6/25 (24%) targets have 1.25 1.25

Hα FWHM > 500 km s−1 while only 2/136 (0.015%)

W1-W2
1.00
1.00

of the controls have Hα FWHM > 500 km s−1 . Broad 0.75

0.75
Balmer lines could indicate the presence of AGN ac- 0.50
0.50
tivity, but there could be potential contamination from 0.25
Jarett+2011
star formation. We did not find any broad outflowing 0.00
Variable
0.25

Non-variable
component in the forbidden [O III] lines of both sam- 1.5 2.0 2.5 3.0 3.5 4.0 4.5 5.0 5.5
0.00
0 10
W2-W3
ples, indicating that neither sample has strong outflows
in ionized gas.
This higher incidence of optical broad lines in the
sample of galaxies exhibiting MIR variability as com- Figure 14. The W ISE color-color diagram for the variable and
non-variable galaxies, with histograms plotting the distribution of
pared to non-variable galaxies is surprising. The pres-
the values in W 1 − W 2 (right) and W 2 − W 3 (top). Although
ence of broad lines in the spectrum of AGN is typically both samples have similar W 1 − W 2 colors, non-variable galaxies
attributed to an orientation effect. In type 1 AGNs, have redder W 2 − W 3 colors.
the nucleus is oriented so that we can observe the BLR
and narrow line regions (NLR). MIR variability, on the bluer W 2 − W 3 colors, indicating that there is likely
other hand, should have no dependence on orientation, lesser contribution from dust heated by star formation
as the variability is believed to originate from the dust in the MIR for the variable targets. This difference in
surrounding the accretion disk and is expected to be the contribution of SF activity is not apparent from the
isotropic. However, Son et al. (2023) found that low- optical, where both samples have similar SFRs (see Fig.
luminosity type 2 AGNs tend to have lower variability 11). Even with similar W 1 − W 2 colors, variable dwarfs
amplitudes than their type 1 counterparts. They at- still have bluer W 2 − W 3 colors. Thus, it is likely that
tribute this to the type 2 AGN having distinctive cen- AGN activity causes redder W 1 − W 2 colors in variable
tral structures due to their low luminosity or their MIR targets, while for the non-variable galaxies, the contri-
brightness being contaminated by emission from the cold bution stems from strong star formation activity.
dust in the host galaxy. On the other hand, recent obser-
vations of extended (tens to hundreds of parsecs) MIR 5. CONCLUSIONS
emission from the polar regions of AGNs suggest that
MIR emission may be dominated by dust from the polar In this work, we selected a sample of twenty-five dwarf
regions rather than from the torus (Asmus et al. 2016; galaxies that are identified to be variable in the MIR
Asmus 2019). This interpretation possibly suggests that from the light curves from the most recent release of the
we can simultaneously observe the BLR as well as MIR AllW ISE/NEOW ISE catalog. We determined that
emission. Our observations of a large number of Type 12/25 (48%) of these galaxies had been confirmed as
1 AGN that are variable in MIR can be explained with AGN in the literature by several optical diagnostic meth-
either of these scenarios. ods, such as the BPT diagram, presence of broad lines
We also obtained W ISE W 1, W 2, W 3, and W 4 fluxes or strong He II emission, indicating that MIR variabil-
for all the galaxies. Examining the infrared colors of ity accurately traces black hole activity in these small
both samples (Fig. 14), we found that non-variable tar- dwarf galaxies. For a subsample of the variable dwarf
gets have higher values of W 2 − W 3 (redder W 2 − W 3 galaxies that did not have any optical confirmations
colors) on average than variable targets (3.92 vs 3.18). of AGN activity, we obtained NIR observations using
Redder W 2−W 3 colors indicate activity caused by star- Keck/NIRES. Our main results can be summarized as
burst events (Jarrett et al. 2013). A p-value of 0.027 follows (also see Fig. 9):
(< 0.05) indicates that there is a statistically significant
1. We present the first detection of a NIR coronal
difference in W 2 − W 3 between the variable and con-
line ([S IX]) in J1205, which has the lowest mass
trol samples. This suggests that variable targets have
(log(M∗ ) = 7.5 M⊙ ) among all the objects in our
20

sample. Most commonly used AGN diagnostics, being available from JWST and the upcoming infrared
such as BPT diagrams and He II emission lines, transient survey telescopes such as the Palomar Gattini-
fail to detect the presence of an AGN in J1205, IR (Moore & Kasliwal 2019), harnessing the power of
thus highlighting that MIR variability can be a MIR variability will likely be fruitful in uncovering large
powerful and sometimes the only tool to detect numbers of active black holes in dwarf galaxies.
AGNs in low-mass galaxies.
We thank the anonymous referee for their thoughtful
2. We find evidence of a broad Paα line potentially
feedback and constructive comments that greatly helped
from the BLR in two of the variable targets (J0928
to improve this paper. Partial support for this project
and J1144). These targets lie in the SF region
was provided by the National Science Foundation un-
of the optical BPT diagram, but they lie in the
der grant No. AST 1817233. The data presented herein
composite or AGN regions of the NIR diagnostic
were obtained at the W. M. Keck Observatory, which
diagrams. The black hole masses implied by their
is operated as a scientific partnership among the Cal-
broad lines are consistent with the black hole to
ifornia Institute of Technology, the University of Cali-
stellar mass relation from RV15.
fornia, and the National Aeronautics and Space Admin-
3. Using NIR diagnostic diagrams, which make use istration. The Observatory was made possible by the
of NIR emission line ratios in order to differentiate generous financial support of the W. M. Keck Founda-
between AGN and SF regions in galaxies, we find tion. The authors wish to recognize and acknowledge
that several of the variable targets (4/9) that we the very significant cultural role and reverence that the
observed moved into the AGN/composite region of summit of Maunakea has always had within the indige-
the diagnostic diagram compared to their location nous Hawaiian community. We are most fortunate to
in the SF regions of the optical BPT diagram. have the opportunity to conduct observations from this
mountain.
4. Comparing a sample of variable and non-variable This publication makes use of data products from the
dwarf galaxies with similar stellar mass and Wide-field Infrared Survey Explorer, which is a joint
W ISE W 1 − W 2 colors, we find that variable project of the University of California, Los Angeles, and
dwarfs have bluer W 2 − W 3 colors than non- the Jet Propulsion Laboratory/California Institute of
variable dwarfs. The redder W 1 − W 2 colors of Technology, and NEOW ISE, which is a project of the
non-variable dwarfs are likely because of high star Jet Propulsion Laboratory/ California Institute of Tech-
formation. Thus, variable dwarfs accurately trace nology. W ISE and NEOW ISE are funded by the Na-
AGN activity independent of mid-infrared colors. tional Aeronautics and Space Administration.
SDSS-IV is managed by the Astrophysical Research
5. Variable dwarfs have a higher fraction of galax-
Consortium for the Participating Institutions of the
ies that have broad (> 500 km s−1 ) Balmer (Hβ
SDSS Collaboration. Funding for the Sloan Digital Sky
and Hα) lines than the non-variable dwarfs (24%
Survey IV has been provided by the Alfred P. Sloan
compared to 0.015%). While this could simply in-
Foundation, the U.S. Department of Energy Office of
dicate a higher incidence of AGN, it is remarkable
Science, and the Participating Institutions. SDSS-IV
that such a large fraction would be type 1 AGN.
acknowledges support and resources from the Center
We find independent confirmation of AGN activity in at for High-Performance Computing at the University of
least 68 % out of 25 MIR variable dwarf galaxies, indi- Utah.
cating that MIR variability can be used to accurately
detect AGN in dwarf galaxies. Improving methods to Facilities: W ISE, NEOW ISE, KCWI/Keck, SDSS
detect low-mass active black holes is the first step in Software: astropy (Astropy Collaboration et al.
constraining the black hole occupation fraction in dwarf 2013, 2018), BADASS (Sexton et al. 2021), pPXF,
galaxies, thus determining the evolutionary pathways for (Cappellari & Emsellem 2004), REDSPEC (https://
black hole seeds in the early universe. Our study shows www2.keck.hawaii.edu/inst/nirspec/redspec.html), Sci-
that using indicators in the infrared is crucial to ob- encePlots (Garrett 2021)
taining a more complete census of black holes in dwarf
galaxies. With the increasing amount of infrared data

APPENDIX
 21

Target [Si VI] flux upper limits

(10−17 erg cm−2 s−1 )
J0847 8.12
J0928 100.97
J1032 81.28
J1144 19.47
J1205 5.55
J1320 3.31
J1358 27.54
J1428 0.55
J1612 4.57
Table 4. 3σ Upper limits to the [Si VI] flux measurements for the non-detections of coronal lines in this galaxies observed with
NIRES (See Figure 5)

The non-uniform detection of coronal lines across galaxies is likely due to insufficient observation times in order to
detect the typically faint coronal lines. We present the upper limits for the [Si VI] flux for the nine galaxies observed
with NIRES in this paper (Table 4). The upper limits to the [Si VI] fluxes were determined by estimating the 1σ noise
levels in the region around [Si VI] flux and multiplying it by three to obtain the 3σ upper limit indicated in Table 4.
None of the observed galaxies had [Si VI] detections.

REFERENCES
Annuar, A., Gandhi, P., Alexander, D. M., et al. 2015, ApJ, Begelman, M. C., Volonteri, M., & Rees, M. J. 2006,
815, 36, doi: 10.1088/0004-637X/815/1/36 MNRAS, 370, 289, doi: 10.1111/j.1365-2966.2006.10467.x
Annuar, A., Alexander, D. M., Gandhi, P., et al. 2017, ApJ, Bellm, E. C., Kulkarni, S. R., Graham, M. J., et al. 2019,
836, 165, doi: 10.3847/1538-4357/836/2/165 PASP, 131, 018002, doi: 10.1088/1538-3873/aaecbe
Arcavi, I., Hosseinzadeh, G., Howell, D. A., et al. 2017, Bi, S., Feng, H., & Ho, L. C. 2020, ApJ, 900, 124,
Nature, 551, 64, doi: 10.1038/nature24291 doi: 10.3847/1538-4357/aba761
Asmus, D. 2019, MNRAS, 489, 2177, Blanton, M. R., & Roweis, S. 2007, AJ, 133, 734,
doi: 10.1093/mnras/stz2289 doi: 10.1086/510127
Asmus, D., Hönig, S. F., & Gandhi, P. 2016, ApJ, 822, 109, Bogdán, Á., Goulding, A. D., Natarajan, P., et al. 2024,
doi: 10.3847/0004-637X/822/2/109 Nature Astronomy, 8, 126,
Astropy Collaboration, Robitaille, T. P., Tollerud, E. J., doi: 10.1038/s41550-023-02111-9
et al. 2013, A&A, 558, A33, Bohn, T., Canalizo, G., Satyapal, S., & Pfeifle, R. W. 2020,
doi: 10.1051/0004-6361/201322068 ApJ, 899, 82, doi: 10.3847/1538-4357/aba52c
Astropy Collaboration, Price-Whelan, A. M., Sipőcz, B. M., Bohn, T., Canalizo, G., Veilleux, S., & Liu, W. 2021, ApJ,
et al. 2018, AJ, 156, 123, doi: 10.3847/1538-3881/aabc4f 911, 70, doi: 10.3847/1538-4357/abe70c
Baldassare, V. F., Geha, M., & Greene, J. 2020, ApJ, 896, Bose, S., Deason, A. J., & Frenk, C. S. 2018, ApJ, 863, 123,
10, doi: 10.3847/1538-4357/ab8936 doi: 10.3847/1538-4357/aacbc4
Baldassare, V. F., Reines, A. E., Gallo, E., & Greene, J. E. Brinchmann, J., Charlot, S., White, S. D. M., et al. 2004,
2017, ApJ, 836, 20, doi: 10.3847/1538-4357/836/1/20 MNRAS, 351, 1151,
Baldassare, V. F., Reines, A. E., Gallo, E., et al. 2016, ApJ, doi: 10.1111/j.1365-2966.2004.07881.x
829, 57, doi: 10.3847/0004-637X/829/1/57 Bromm, V., & Loeb, A. 2003, ApJ, 596, 34,
Baldwin, J. A., Phillips, M. M., & Terlevich, R. 1981, doi: 10.1086/377529
PASP, 93, 5, doi: 10.1086/130766 Burke, C. J., Liu, Y., Ward, C. A., et al. 2024, arXiv
Becker, G. D., Rauch, M., & Sargent, W. L. W. 2009, ApJ, e-prints, arXiv:2402.06882,
698, 1010, doi: 10.1088/0004-637X/698/2/1010 doi: 10.48550/arXiv.2402.06882
Beckmann, R. S., Dubois, Y., Volonteri, M., et al. 2023, Burke, C. J., Shen, Y., Chen, Y.-C., et al. 2020, ApJ, 899,
MNRAS, 523, 5610, doi: 10.1093/mnras/stad1544 136, doi: 10.3847/1538-4357/aba3ce
22

Burke, C. J., Shen, Y., Liu, X., et al. 2023, MNRAS, 518, Haidar, H., Habouzit, M., Volonteri, M., et al. 2022,
1880, doi: 10.1093/mnras/stac2478 MNRAS, 514, 4912, doi: 10.1093/mnras/stac1659
Burke, C. J., Shen, Y., Blaes, O., et al. 2021, Science, 373, Hainline, K. N., Reines, A. E., Greene, J. E., & Stern, D.
789, doi: 10.1126/science.abg9933 2016, ApJ, 832, 119, doi: 10.3847/0004-637X/832/2/119
Burke, C. J., Liu, X., Shen, Y., et al. 2022, MNRAS, 516, Harikane, Y., Ouchi, M., Oguri, M., et al. 2023, ApJS, 265,
2736, doi: 10.1093/mnras/stac2262 5, doi: 10.3847/1538-4365/acaaa9
Calabrò, A., Pentericci, L., Feltre, A., et al. 2023, A&A, Harish, S., Malhotra, S., Rhoads, J. E., et al. 2023, ApJ,
679, A80, doi: 10.1051/0004-6361/202347190 945, 157, doi: 10.3847/1538-4357/acb99c
Cann, J. M., Satyapal, S., Abel, N. P., et al. 2019, ApJL, Hinkle, J. T., Tucker, M. A., Shappee, B. J., et al. 2023,
870, L2, doi: 10.3847/2041-8213/aaf88d MNRAS, 519, 2035, doi: 10.1093/mnras/stac3659
Cann, J. M., Satyapal, S., Rothberg, B., et al. 2021, ApJL, Inayoshi, K., Visbal, E., & Haiman, Z. 2020, ARA&A, 58,
912, L2, doi: 10.3847/2041-8213/abf56d 27, doi: 10.1146/annurev-astro-120419-014455
Cappellari, M., & Emsellem, E. 2004, PASP, 116, 138, Ivezić, Ž., Kahn, S. M., Tyson, J. A., et al. 2019, ApJ, 873,
doi: 10.1086/381875 111, doi: 10.3847/1538-4357/ab042c
Chadayammuri, U., Bogdán, Á., Ricarte, A., & Natarajan, Izotov, Y. I., & Thuan, T. X. 2009, ApJ, 707, 1560,
P. 2023, ApJ, 946, 51, doi: 10.3847/1538-4357/acbea6 doi: 10.1088/0004-637X/707/2/1560
Chilingarian, I. V., Katkov, I. Y., Zolotukhin, I. Y., et al. Izotov, Y. I., Thuan, T. X., & Guseva, N. G. 2017,
MNRAS, 471, 548, doi: 10.1093/mnras/stx1629
2018, ApJ, 863, 1, doi: 10.3847/1538-4357/aad184
—. 2021, MNRAS, 508, 2556, doi: 10.1093/mnras/stab2798
Clark, P., Graur, O., Callow, J., et al. 2024, MNRAS, 528,
Jakobsen, P., Ferruit, P., Alves de Oliveira, C., et al. 2022,
7076, doi: 10.1093/mnras/stae460
A&A, 661, A80, doi: 10.1051/0004-6361/202142663
Colina, L., Piqueras López, J., Arribas, S., et al. 2015,
Jarrett, T. H., Masci, F., Tsai, C. W., et al. 2013, AJ, 145,
A&A, 578, A48, doi: 10.1051/0004-6361/201425567
6, doi: 10.1088/0004-6256/145/1/6
Dark Energy Survey Collaboration, Abbott, T., Abdalla,
Jones, T., Martin, C., & Cooper, M. C. 2015, ApJ, 813,
F. B., et al. 2016, MNRAS, 460, 1270,
126, doi: 10.1088/0004-637X/813/2/126
doi: 10.1093/mnras/stw641
Karambelkar, V. R., Adams, S. M., Whitelock, P. A., et al.
De, K., MacLeod, M., Karambelkar, V., et al. 2023, Nature,
2019, ApJ, 877, 110, doi: 10.3847/1538-4357/ab1a41
617, 55, doi: 10.1038/s41586-023-05842-x
Kauffmann, G., Heckman, T. M., White, S. D. M., et al.
den Brok, J. S., Koss, M. J., Trakhtenbrot, B., et al. 2022,
2003a, MNRAS, 341, 33,
ApJS, 261, 7, doi: 10.3847/1538-4365/ac5b66
doi: 10.1046/j.1365-8711.2003.06291.x
den Brok, M., Seth, A. C., Barth, A. J., et al. 2015, ApJ,
Kauffmann, G., Heckman, T. M., Tremonti, C., et al.
809, 101, doi: 10.1088/0004-637X/809/1/101
2003b, MNRAS, 346, 1055,
Devecchi, B., Volonteri, M., Rossi, E. M., Colpi, M., &
doi: 10.1111/j.1365-2966.2003.07154.x
Portegies Zwart, S. 2012, MNRAS, 421, 1465,
Kawaguchi, T., & Mori, M. 2011, ApJ, 737, 105,
doi: 10.1111/j.1365-2966.2012.20406.x doi: 10.1088/0004-637X/737/2/105
Do, T., Wright, S. A., Barth, A. J., et al. 2014, AJ, 147, 93, Kelson, D. D. 2003, PASP, 115, 688, doi: 10.1086/375502
doi: 10.1088/0004-6256/147/4/93 Kennicutt, Robert C., J. 1998, ARA&A, 36, 189,
Donley, J. L., Rieke, G. H., Pérez-González, P. G., & doi: 10.1146/annurev.astro.36.1.189
Barro, G. 2008, ApJ, 687, 111, doi: 10.1086/591510 Kewley, L. J., & Dopita, M. A. 2002, ApJS, 142, 35,
Finkelstein, S. L., Bagley, M. B., Ferguson, H. C., et al. doi: 10.1086/341326
2023, ApJL, 946, L13, doi: 10.3847/2041-8213/acade4 Kewley, L. J., & Ellison, S. L. 2008, ApJ, 681, 1183,
Garrett, J. 2021, doi: 10.5281/zenodo.5512926 doi: 10.1086/587500
Goulding, A. D., Greene, J. E., Setton, D. J., et al. 2023, Kewley, L. J., Geller, M. J., & Jansen, R. A. 2004, AJ, 127,
ApJL, 955, L24, doi: 10.3847/2041-8213/acf7c5 2002, doi: 10.1086/382723
Greene, J. E. 2012, Nature Communications, 3, 1304, Kim, D., Im, M., Canalizo, G., et al. 2018, ApJS, 238, 37,
doi: 10.1038/ncomms2314 doi: 10.3847/1538-4365/aadfd5
Greene, J. E., & Ho, L. C. 2007, ApJ, 670, 92, Kimura, Y., Yamada, T., Kokubo, M., et al. 2020, ApJ,
doi: 10.1086/522082 894, 24, doi: 10.3847/1538-4357/ab83f3
Greene, J. E., Strader, J., & Ho, L. C. 2020, ARA&A, 58, Kocevski, D. D., Onoue, M., Inayoshi, K., et al. 2023,
257, doi: 10.1146/annurev-astro-032620-021835 ApJL, 954, L4, doi: 10.3847/2041-8213/ace5a0
 23

Kokorev, V., Fujimoto, S., Labbe, I., et al. 2023, ApJL, Miller, B. P., Gallo, E., Greene, J. E., et al. 2015, ApJ, 799,
957, L7, doi: 10.3847/2041-8213/ad037a 98, doi: 10.1088/0004-637X/799/1/98
Lamperti, I., Koss, M., Trakhtenbrot, B., et al. 2017, Moore, A. M., & Kasliwal, M. M. 2019, Nature Astronomy,
MNRAS, 467, 540, doi: 10.1093/mnras/stx055 3, 109, doi: 10.1038/s41550-018-0675-x
Landt, H., Bentz, M. C., Ward, M. J., et al. 2008, ApJS, Moran, E. C., Shahinyan, K., Sugarman, H. R., Vélez,
174, 282, doi: 10.1086/522373 D. O., & Eracleous, M. 2014, AJ, 148, 136,
Larkin, J. E., Armus, L., Knop, R. A., Soifer, B. T., & doi: 10.1088/0004-6256/148/6/136
Matthews, K. 1998, ApJS, 114, 59, doi: 10.1086/313063 Müller-Sánchez, F., Hicks, E. K. S., Malkan, M., et al.
Law, N. M., Kulkarni, S. R., Dekany, R. G., et al. 2009, 2018, ApJ, 858, 48, doi: 10.3847/1538-4357/aab9ad
PASP, 121, 1395, doi: 10.1086/648598
Natarajan, P. 2014, General Relativity and Gravitation, 46,
Lequeux, J., Peimbert, M., Rayo, J. F., Serrano, A., &
1702, doi: 10.1007/s10714-014-1702-6
Torres-Peimbert, S. 1979, A&A, 80, 155
—. 2021, MNRAS, 501, 1413, doi: 10.1093/mnras/staa3724
Li, J., & Shen, Y. 2023, ApJ, 950, 122,
Nguyen, D. D., Seth, A. C., Neumayer, N., et al. 2018, ApJ,
doi: 10.3847/1538-4357/accade
858, 118, doi: 10.3847/1538-4357/aabe28
Lodato, G., & Natarajan, P. 2006, MNRAS, 371, 1813,
—. 2019, ApJ, 872, 104, doi: 10.3847/1538-4357/aafe7a
doi: 10.1111/j.1365-2966.2006.10801.x
Lupi, A., Haardt, F., Dotti, M., et al. 2016, MNRAS, 456, Oh, K., Yi, S. K., Schawinski, K., et al. 2015, ApJS, 219, 1,
2993, doi: 10.1093/mnras/stv2877 doi: 10.1088/0067-0049/219/1/1
Lyke, B. W., Higley, A. N., McLane, J. N., et al. 2020, Ohlson, D., Seth, A. C., Gallo, E., Baldassare, V. F., &
ApJS, 250, 8, doi: 10.3847/1538-4365/aba623 Greene, J. E. 2024, AJ, 167, 31,
MacLeod, C. L., Ivezić, Ž., Kochanek, C. S., et al. 2010, doi: 10.3847/1538-3881/acf7bc
ApJ, 721, 1014, doi: 10.1088/0004-637X/721/2/1014 Pace, Z. J., Tremonti, C., Chen, Y., et al. 2019, ApJ, 883,
Madau, P., & Rees, M. J. 2001, ApJL, 551, L27, 83, doi: 10.3847/1538-4357/ab3726
doi: 10.1086/319848 Pagel, B. E. J., Edmunds, M. G., Blackwell, D. E., Chun,
Maiolino, R., Nagao, T., Grazian, A., et al. 2008, A&A, M. S., & Smith, G. 1979, MNRAS, 189, 95,
488, 463, doi: 10.1051/0004-6361:200809678 doi: 10.1093/mnras/189.1.95
Maiolino, R., Scholtz, J., Curtis-Lake, E., et al. 2023, arXiv Paudel, S., Hilker, M., Ree, C. H., & Kim, M. 2016, ApJL,
e-prints, arXiv:2308.01230, 820, L19, doi: 10.3847/2041-8205/820/1/L19
doi: 10.48550/arXiv.2308.01230 Petter, G. C., Hickox, R. C., Alexander, D. M., et al. 2023,
Maraston, C., Daddi, E., Renzini, A., et al. 2006, ApJ, 652, ApJ, 946, 27, doi: 10.3847/1538-4357/acb7ef
85, doi: 10.1086/508143 Pettini, M., & Pagel, B. E. J. 2004, MNRAS, 348, L59,
Marleau, F. R., Clancy, D., Habas, R., & Bianconi, M. doi: 10.1111/j.1365-2966.2004.07591.x
2017, A&A, 602, A28, doi: 10.1051/0004-6361/201629832 Polimera, M. S., Kannappan, S. J., Richardson, C. T., et al.
Martı́nez-Palomera, J., Lira, P., Bhalla-Ladd, I., Förster, 2022, ApJ, 931, 44, doi: 10.3847/1538-4357/ac6595
F., & Plotkin, R. M. 2020, ApJ, 889, 113,
Portegies Zwart, S. F., Baumgardt, H., Hut, P., Makino, J.,
doi: 10.3847/1538-4357/ab5f5b
& McMillan, S. L. W. 2004, Nature, 428, 724,
Martı́nez-Palomera, J., Förster, F., Protopapas, P., et al.
doi: 10.1038/nature02448
2018, AJ, 156, 186, doi: 10.3847/1538-3881/aadfd8
Reines, A. E. 2022, Nature Astronomy, 6, 26,
Martins, L. P., Rodrı́guez-Ardila, A., Diniz, S., Gruenwald,
doi: 10.1038/s41550-021-01556-0
R., & de Souza, R. 2013, MNRAS, 431, 1823,
Reines, A. E., Greene, J. E., & Geha, M. 2013, ApJ, 775,
doi: 10.1093/mnras/stt296
116, doi: 10.1088/0004-637X/775/2/116
Mason, R. E., Rodrı́guez-Ardila, A., Martins, L., et al.
2015, ApJS, 217, 13, doi: 10.1088/0067-0049/217/1/13 Reines, A. E., & Volonteri, M. 2015, ApJ, 813, 82,
Mezcua, M. 2017, International Journal of Modern Physics doi: 10.1088/0004-637X/813/2/82
D, 26, 1730021, doi: 10.1142/S021827181730021X Rho, J., Geballe, T. R., Banerjee, D. P. K., et al. 2018,
Mezcua, M., Civano, F., Marchesi, S., et al. 2018, MNRAS, ApJL, 864, L20, doi: 10.3847/2041-8213/aad77f
478, 2576, doi: 10.1093/mnras/sty1163 Ricci, C., Bauer, F. E., Arevalo, P., et al. 2016, ApJ, 820, 5,
Mezcua, M., & Domı́nguez Sánchez, H. 2020, ApJL, 898, doi: 10.3847/0004-637X/820/1/5
L30, doi: 10.3847/2041-8213/aba199 Riffel, R., Rodrı́guez-Ardila, A., & Pastoriza, M. G. 2006,
—. 2024, MNRAS, 528, 5252, doi: 10.1093/mnras/stae292 A&A, 457, 61, doi: 10.1051/0004-6361:20065291
24

Rodrı́guez-Ardila, A., Prieto, M. A., Portilla, J. G., & Ulrich, M.-H., Maraschi, L., & Urry, C. M. 1997, ARA&A,
Tejeiro, J. M. 2011, ApJ, 743, 100, 35, 445, doi: 10.1146/annurev.astro.35.1.445
doi: 10.1088/0004-637X/743/2/100 van Wassenhove, S., Volonteri, M., Walker, M. G., & Gair,
Rosswog, S. 2005, ApJ, 634, 1202, doi: 10.1086/497062 J. R. 2010, MNRAS, 408, 1139,
Ruan, J. J., Anderson, S. F., Plotkin, R. M., et al. 2014, doi: 10.1111/j.1365-2966.2010.17189.x
ApJ, 797, 19, doi: 10.1088/0004-637X/797/1/19 Vanden Berk, D. E., Wilhite, B. C., Kron, R. G., et al.
Salehirad, S., Reines, A. E., & Molina, M. 2022, ApJ, 937,
2004, ApJ, 601, 692, doi: 10.1086/380563
7, doi: 10.3847/1538-4357/ac8876
Vazdekis, A., Koleva, M., Ricciardelli, E., Röck, B., &
Salim, S., Rich, R. M., Charlot, S., et al. 2007, ApJS, 173,
Falcón-Barroso, J. 2016, MNRAS, 463, 3409,
267, doi: 10.1086/519218
doi: 10.1093/mnras/stw2231
Sartori, L. F., Schawinski, K., Treister, E., et al. 2015,
MNRAS, 454, 3722, doi: 10.1093/mnras/stv2238 Veilleux, S., Goodrich, R. W., & Hill, G. J. 1997, ApJ, 477,
Secrest, N. J., & Satyapal, S. 2020, ApJ, 900, 56, 631, doi: 10.1086/303735
doi: 10.3847/1538-4357/ab9309 Veilleux, S., & Osterbrock, D. E. 1987, ApJS, 63, 295,
Sesar, B., Ivezić, Ž., Lupton, R. H., et al. 2007, AJ, 134, doi: 10.1086/191166
2236, doi: 10.1086/521819 Veron-Cetty, M. P., & Veron, P. 2006, VizieR Online Data
Sexton, R. O., Matzko, W., Darden, N., Canalizo, G., & Catalog: Quasars and Active Galactic Nuclei (12th Ed.),
Gorjian, V. 2021, MNRAS, 500, 2871, CDS/ADC Collection of Electronic Catalogues, Vol.
doi: 10.1093/mnras/staa3278 7248, p. 0-1 (2006)
Shin, L., Woo, J.-H., Son, D., et al. 2022, AJ, 163, 73, Volonteri, M., & Begelman, M. C. 2010, MNRAS, 409,
doi: 10.3847/1538-3881/ac4038 1022, doi: 10.1111/j.1365-2966.2010.17359.x
Shirazi, M., & Brinchmann, J. 2012, MNRAS, 421, 1043,
Volonteri, M., Habouzit, M., & Colpi, M. 2021, Nature
doi: 10.1111/j.1365-2966.2012.20439.x
Reviews Physics, 3, 732, doi: 10.1038/s42254-021-00364-9
Silk, J., & Rees, M. J. 1998, A&A, 331, L1,
Volonteri, M., & Rees, M. J. 2005, ApJ, 633, 624,
doi: 10.48550/arXiv.astro-ph/9801013
doi: 10.1086/466521
Smith, K. L., Koss, M., & Mushotzky, R. F. 2014, ApJ,
794, 112, doi: 10.1088/0004-637X/794/2/112 Ward, C., Gezari, S., Nugent, P., et al. 2022, ApJ, 936, 104,
Smitka, M. T. 2016, PhD thesis, Texas A&M University doi: 10.3847/1538-4357/ac8666
System Wasleske, E. J., & Baldassare, V. F. 2024, arXiv e-prints,
Son, S., Kim, M., & Ho, L. C. 2022, ApJ, 927, 107, arXiv:2405.20312, doi: 10.48550/arXiv.2405.20312
doi: 10.3847/1538-4357/ac4dfc Wilson, J. C., Henderson, C. P., Herter, T. L., et al. 2004,
—. 2023, ApJ, 958, 135, doi: 10.3847/1538-4357/ad01bc in Society of Photo-Optical Instrumentation Engineers
Soraisam, M. D., Szalai, T., Van Dyk, S. D., et al. 2023, (SPIE) Conference Series, Vol. 5492, Ground-based
ApJ, 957, 64, doi: 10.3847/1538-4357/acef22 Instrumentation for Astronomy, ed. A. F. M. Moorwood
Sun, L., Xiao, L., & Li, G. 2022, MNRAS, 513, 4057, & M. Iye, 1295–1305, doi: 10.1117/12.550925
doi: 10.1093/mnras/stac1121 Wright, E. L., Eisenhardt, P. R. M., Mainzer, A. K., et al.
Treiber, H. P., Hinkle, J. T., Fausnaugh, M. M., et al. 2023, 2010, AJ, 140, 1868, doi: 10.1088/0004-6256/140/6/1868
MNRAS, 525, 5795, doi: 10.1093/mnras/stad2530
York, D. G., Adelman, J., Anderson, John E., J., et al.
Übler, H., Maiolino, R., Curtis-Lake, E., et al. 2023, A&A,
2000, AJ, 120, 1579, doi: 10.1086/301513
677, A145, doi: 10.1051/0004-6361/202346137