Gaia17bpp: A Giant Star with the Deepest and Longest Known Dimming Event

We report the serendipitous discovery of Gaia17bpp/2MASS J19372316+1759029, a binary star with a deep single large-amplitude dimming event of ∼4.5 mag that lasted over 6.5 yr. Using the optical-to-IR spectral energy distribution (SED), we constrain the primary star to be a cool giant M0III star with effective temperature T eff = 3850 K and radius R = 58 R ⊙. Based on the SED fitting, we obtained a bimodal posterior distribution of primary stellar masses with a stronger preference for a 1.5 M ⊙ mass star. Within the last 66 yr of photometric coverage, no other significant dimming events of this depth and duration were identified in the optical light curves. Using a Gaussian process, we fit a generalized Gaussian distribution to the optical and IR light curves and conclude that the dimming event exhibits moderate asymmetries from optical to IR. At the minimum of the dimming event, the Wide-field Infrared Survey Explorer color (W1–W2) differed by ∼0.2 mag relative to the primary star color outside the dimming event. The ingress and egress colors show a shallow reddening profile. We suggest that the main culprit of the dimming event is likely due to the presence of a large, optically thick disk transiting the primary giant star. By fitting a monochromatic transit model of an oblate disk transiting a star, we found good agreement with a slow-moving (0.005 km s−1) disk with a ∼1.4 au radius. We propose that Gaia17bpp belongs to a rare binary star population similar to the ϵ Aurigae system, which consists of a secondary star enshrouded by an optically thick debris disk.


Introduction
The long-term photometric monitoring of the sky is steadily increasing the discovery of remarkable dimming events on the timescale of decades. Among slowly evolving stellar variables, there is an emerging population of photometrically deep, longduration, and long-period anomalous binary systems with the secondary star being enveloped by a large disk that calls for more attention (Stencel 2016). At the turn of the century, the enigmatic star ò Aurigae (ò Aur), which exhibits a single flatbottomed, ∼2 yr long dimming event, challenged our understanding of binary stars (Carroll et al. 1991). Today, more than a few decades later, ò Aur has been understood to be a lowmass F0 supergiant star surrounded by a young B5V-type star companion with an extended cool disk with an orbital period of 27 yr. Both spectroscopic and direct interferometric imaging techniques have confirmed the presence of a geometrically thin and optically thick disk that is consistent with debris (Kloppenborg et al. 2010;Stencel et al. 2011). Open questions such as the origins and lifetime of the disk, occurrence rates, and formation channels of such systems remain unexplored. Gibson & Stencel (2018) conducted a theoretical study using MESA modules to explain the evolution of the ò Aur binary system. Their study suggests that the formation of the large debris disk found around the secondary star of ò Aur was the product of accretion from two intermediate-mass stars, 9.85 and 4.5 M e , with an initial orbital period of roughly 100 days, after the primary star had made its first ascent to the post-red giant branch/pre-asymptotic giant branch phase. As noted, however, their models could not replicate the large orbital period of ò Aur, which was likely due to poor constraints on the mass transfer profile.
Prior to this work, TYC 2505-672-1 (Lipunov et al. 2016) previously held the record with the longest-duration dimming event and longest orbital period binary. The system consists of an M2III giant that undergoes a 4.5 mag 3.5 yr long deep eclipse every 69.1 yr (Rodriguez et al. 2016). It is suggested that the companion star was a cool subdwarf B (sdB) type star that will eventually become a low-mass pre-helium white dwarf (WD), enshrouded by an opaque disk. The origin of the disk, however, remains puzzling since accretion from the primary would be challenging to explain given its large semimajor-axis separation (Rodriguez et al. 2016). Additionally, previous searches for pre-helium WD stars have been found in compact binary systems (Maxted et al. 2014;van Roestel et al. 2018), unlike a very long binary such as TYC 2505-672-1. More recently, reminiscent systems such as VVV-WIT-08 (Smith et al. 2021), a K7 giant star that showed a single smooth dimming event, have also been thought to be occulted by an optically thick disk. Despite the limited understanding of the occulting disk in the majority of these dimming stars, a significant number of them exhibit similar characteristics. These include an evolved primary star, prolonged periods of single deep dimming events that last for years, large orbital periods, and the absence of IR excess. Finally, other identified long-period binaries have also claimed the presence of an extended disk around the companion star, such as ASASSN-21co (Rowan et al. 2021) and η Gem (Torres & Sakano 2022), that causes prolonged deep eclipses.
In this study, we report the serendipitous discovery of a new giant dimming star, Gaia17bpp/2MASS J19372316+175902, which now holds the record for the longest and deepest dimming event. We incorporate the use of multiwavelength light curves and optical spectra of this star to understand the nature of the dimming mechanism. This paper is organized as follows. In Section 2 we discuss the data acquisition from allsky surveys and spectroscopic follow-up efforts. Section 3 describes the modeling of the spectral energy distribution (SED) of the primary star using multiband archival photometry. Similarly, in Section 4 we analyze the properties and shapes of the light curves from the optical to infrared wavelengths. We provide an interpretation of the dimming event, fit a monochromatic oblate disk model using IR light curves, and investigate the possible mechanism driving it in Section 5. Finally, in Section 6 we provide our conclusions.

Data
In this section, we describe the data acquisition from multiple public all-sky surveys. All collated time-series photometric measurements have been combined to generate a mosaic light curve of Gaia17bpp spanning from the late 1950s to the present.

Gaia Photometric Science Alerts
Gaia17bpp was first saved as an alert through the Gaia photometric alert stream (Delgado et al. 2017) and was discovered through a subsequent search for deep and long (>5 yr) stellar variability. Gaia has been scanning the night sky since 2014 and has been collecting photometric measurements in the Gaia G bandpass and releasing alerts through the Gaia Photometric Science Alerts (GPSA) described in Hodgkin et al. (2021). We query the Gaia alert light curves through the Gaia Alert website portal 1 to collect all the available epochal Gaia G, BP, and RP detections. In Figure 1 we show the issued alert Gaia G light curve of Gaia17bpp.
We note that the current alert schema does not include photometric uncertainties; however, Hodgkin et al. (2021) suggest an empirical means of estimating uncertainties as a function of source brightness. We also mined 2 the uncalibrated BP/RP flux spectra following the approach of Hodgkin et al. (2021), using the integrated BP/RP spectra as a reasonable estimate of the color evolution in the optical regime. The BP photometry in this case suffers from higher scatter because the baseline BP photometry is close to the limiting magnitude. To examine the difference between the estimated uncalibrated BP and RP photometry, we cross-matched all photometric alerts that were also found in the Gaia Data Release 3 (Gaia Collaboration et al. 2023) and compared the reported BP and RP magnitudes with the ones derived from this work. On average, we found a standard deviation of 1.5 mag per magnitude bin. For our reported BP/RP detections, we inflated the photometric errors to capture any excess uncertainty by 1.5σ G .

Gaia Data Releases
According to the full photometric history enabled through the GPSA, Gaia17bpp was already at its faintest phase and close to the detection threshold of Gaia when observations began. We identified the system in both Gaia DR2 and DR3 catalogs and investigated the astrometric solutions, specifically the parallax and proper-motion vectors. In both data releases the parallax is negative. We suspect that this is due to the photometric uncertainty at the bottom of the dimming event, leading to a negative parallax solution (Luri et al. 2018). To obtain distances, we rely on the photogeometric distances derived from the work of Bailer-Jones et al. (2021) that compensates for negative parallax while enabling a strong prior from the available photometry and Galactic extinction at the line of sight. We adopt the median photogeometric posterior distribution distance of -+ 8.5 1.6 2.3 kpc throughout all calculations in this study. In the presence of negative parallaxes, we caution that this estimated distance is mostly prior dominated. In Table 1 we highlight all the key parameters reported from Gaia DR3.
Enabled through the distance estimate and extrapolation of BP/RP magnitudes, we placed the location of Gaia17bpp on the observed Hertzsprung−Russell diagram at the present date as seen in Figure 2. We also attempted to track the evolution of Gaia17bpp throughout its dimming phase. We bin both the BP/RP and G-band magnitudes by a 450-day running median  to capture the overall color and absolute magnitude evolution. In general, we found consistency in the reported photometry with both the official Gaia data releases and the photometric science alerts. However, we noticed that the dimming event was predominantly bluer than the quiescent star. As noted by Riello et al. (2021), faint sources with red colors tend to have overestimated BP fluxes. To mitigate the effects of overestimating the BP flux, we only computed the BP/RP magnitudes based on the epochal spectra after 2016 when the source began rebrightening.

WISE
Subsequent investigation revealed that Gaia17bpp was also well monitored by the Wide-field Infrared Survey Explorer (WISE; Wright et al. 2010;Mainzer et al. 2014) and NEOWISE (Mainzer et al. 2011) in the W1-W4 filters. To query the WISE and NEOWISE light curves, we use an opensource Python tool, WISE Light Curves, 3 that automatically queries the NASA/IPAC Infrared Science Archive to compile the epochal photometry from both surveys. We use a 5″ cone search query and identify WISE J193723.16 + 175903, confirming that it is the identical source to Gaia17bpp. We also searched through the AKARI and Planck observations and found no excess flux within 1 deg 2 in the far-infrared.

MAST
We also queried the Barbara A. Mikulski Archive for Space Telescopes (MAST) data archive and found a few optical observations from the Panoramic Survey Telescope and Rapid Response System (PS1; Chambers et al. 2016), two near-UV (NUV) images from the Galaxy Evolution Explorer (GALEX; Martin et al. 2005) in the All-Sky Imaging Survey (AIS), and more recently three full-frame images from the Transiting Exoplanet Survey Satellite (TESS; Ricker et al. 2014). We generated the PS1 light curves using the Space Telescope Science Institute database to query the PS1 Detection table fields 4 with a cross-match cone radius of 3″. The PS1 detection table contains single-epoch photometry from single-visit exposure that produces the point-spread function aperture photometry of detected sources. Both GALEX NUV raw count images were visually inspected; however, no coincident source at the location of the Gaia17bpp system was found. We also searched the GALEX GR6/GR7 sources and only found the nearest source to be 1 4 away, making it unlikely that this is our target. Based on the GALEX AIS 20.5 mag NUV limiting magnitude, we conclude that Gaia17bpp must be fainter.

Archival Photometry
We obtained archival photometry using the Carlsberg Meridian Catalogue 15 (CMC15;  Walton et al. 2004), and Digital Access to a Sky Century at Harvard (DASCH; Grindlay 2017). The DASCH photographic plates, while having been scanned, had not reported observed magnitudes and extractions at the location of the star. Instead, we ran a source extraction tool to perform aperture photometry at the location of Gaia17bpp using SEP (Barbary 2018). For each 400″ × 400″ scanned plate from DASCH, we cross-matched each 2σ source detection to the PS1 r-band catalog assuming a limiting magnitude of 16 mag. From each science image, we subtract out the background rms variation. After comparing and extrapolating the zero-point calibration between the PS1 and DASCH images, we summed the fixed aperture at the location of Gaia17bpp to extrapolate a rough estimate of its magnitude and error. 5 Outside of the primary dimming event, Gaia17bpp was also detected in the Zwicky Transient Facility (ZTF) survey  in the gri broadband filters. In short, ZTF is a 47 deg 2 time-domain survey utilizing the 48-inch Schmidt Palomar telescope (P48) located at the Palomar Observatory. ZTF performs a public component survey that covers the entire northern sky with a 2-day cadence in g + r (≈34% of P48 time) and a 1-day Galactic plane survey in g + r (≈6% of P48 time). A high-cadence ≈2500 deg 2 survey (3g + 3r per night) was conducted for collaboration time, including a 4-day i-band survey. On average the survey covers ∼3750 deg 2 hr −1 with a median limiting magnitude of r ∼ 20.6 mag, g ∼ 20.8 mag, and i ∼ 19.9 mag 5σ limiting magnitude in 30 s exposures (Dekany et al. 2020). For in-depth coverage of the scientific goals and survey strategy, we suggest that the reader review Bellm et al. (2019) and Graham et al. (2019). No other ZTF or GPSA has been issued within 20″ of Gaia17bpp. In Figure 3, we show the optical-to-IR mosaic light curve of Gaia17bpp.

Spectroscopy
Gaia17bpp was observed with the Apache Point Observatory (APO) ARC 3.5 m telescope using a medium-dispersion spectrograph KOSMOS (Stoll et al. 2010) on 2022 September 18. We used the red high slit configuration with wavelength coverage between 6150 and 9800 Å. Our on-sky time included one 30-minute exposure with clear sky conditions. Our data reduction followed standard long-slit spectroscopy procedures and reduced the optical spectrum using the open-source Python package PyKOSMOS (Davenport 2021). In Figure 4, we show the final reduced optical spectrum of Gaia17bpp without correcting for the telluric absorption features. In the same figure, we also include the near-IR spectra of an M0 dwarf and giant, HD 19305 (M0V) and HD 213893 (M0IIIb), standard stars (Rayner et al. 2009). Upon inspecting the optical spectrum of Gaia17bpp, we do not find any immediate anomalous spectroscopic features. Generally, we identified several absorption lines, such as a prominent Ca II triplet at 8498, 8542, and 8662 Å, and clear 7050 Å TiO bands,  including other molecular lines near 8160 Å. We did not identify other significant narrow emission lines that appear in the spectrum, except for a very weak Hα emission line. The absence of the calcium hydride (CaH 2 ) λλ6382, 6830, Na I doublet λλ8183, 8195, and weak K I λλ7665, 7699 lines indicate the presence of an evolved giant star (Stephenson 1986;Allers et al. 2007). The identified spectroscopic features are generally consistent with an ordinary late-type cool K-to-Mclass star. In Section 3 we discuss in depth how the interpretation of such absorption lines is consistent with the classification of an RGB star as the primary source. The collected spectrum is available for further inspection on the Transient Naming Server (TNS) under the source name AT2017exj. 6

Follow-up Photometry
On 2023 May 7 we performed follow-up photometry on Gaia17bpp with the APO 3.5 m Astrophysical Research Consortium Telescope Imaging Camera (ARCTIC; Huehnerhoff et al. 2016) using the Sloan Digital Sky Survey (SDSS) u filter. The seeing was approximately 1″ throughout the entire night. We note that the Moon was ∼91% full, with a large >80°separation from our target. We observed the Gaia17bpp for a total of 23 3-minute images to eliminate sky brightness background and noise levels. We did not perform dithering. We reduced our data following standard calibration procedures. We assumed a photometric zero-point offset for the SDSS u filter of 24 mag. We used the Python open-source tool photutils (Bradley et al. 2022) to perform the standard reduction of the SDSS u-band science images including our calibration files. No source was detected within 10″ of the Gaia17bpp sky position. Based on the 5σ identified sources we found in our deep ARCTIC images, the distribution was roughly Gaussian with an average SDSS u-band magnitude of ∼22 mag.

Spectral Energy Distribution
We attempt to infer the physical parameters of the primary star. Given the rich multiwavelength archival photometry, we perform a stellar SED modeling to constrain the properties of the primary star using open-source SED fitter Ariadne (Vines & Jenkins 2022). Ariadne deploys a suite of stellar atmosphere model grids by performing a convolution of synthetic SED atmospheres and broadband photometry to constrain the effective temperature, metallicity, distance, radius, and extinction of a given SED. It performs a dynamic Bayesian multinested sampling algorithm through dynesty (Speagle 2020) and amalgamates each stellar atmospheric model using a Bayesian model averaging (BMA) to finally estimate the best-fitted model parameters of a given primary star through a weighted average. In our case, we began by including all the observed broadband photometry outside the dimming event. In Table 2 we summarize the median observed magnitudes outside the primary dimming event from all epochs before 2013 or after 2018. For the fitting, we assumed a Fitzpatrick (1999) extinction law with A V = 5.3 assuming a Galactic R V = 3.1 value. In Table 3 we show the priors and posterior distributions used for this analysis.
In Figure 5 we show the synthetic BMA PHOENIX atmospheric model that best fits our broadband photometry up to 4.6 μm. In our SED plot, we also include the W3 detection and W4 upper limit of the primary star; however, the SED fitter does not fit those detections. We extended the synthetic PHOENIX model continuum by fitting a blackbody function at the best-fitted temperature, radius, and distance (shown with dashed lines in Figure 5). The average fitted model shows convergence on well-constrained posterior distribution seen in Figure 6, including their 25th, 50th, and 75th percentiles. The posterior distance distribution is in good agreement with the photogeometric distance estimation. According to the physical estimates we derive from the SED fitting, we find the primary star to be consistent with a K5III or M0III giant star according to the classification regime considering its effective temperature, radius, and surface gravity (Dyck et al. 1996;Dumm & Schild 1998;Ceillier et al. 2017). The interpretation of a cool giant star is also consistent with the identified spectroscopic features discussed in Section 2.5. Using the maximum likelihood from the BMA, Ariadne also fits in the background MIST isochrones given the derived luminosity and temperature to estimate the mass. In Figure 7 we show the posterior distribution of the resulting stellar mass, including the kernel density estimation (KDE). It is evident that the isochrone mass posterior distribution is bimodal with a higher probability density for a lower-mass stellar model with a median of 1.56 M e . Note. We assume that the intrinsic stellar SED is the same before and after the dimming event. a 2σ upper limit.

Primary Star Variability
Since the primary large-amplitude dimming event, the Gaia17bpp system thus far has not exhibited any other largescale variability beyond an overall 0.05 mag baseline scatter that was calculated using the ZTF-g, ZTF-r, and Gaia G light curve at all observations after 2018. We searched the ZTF-g, ZTF-r, and Gaia G light curves for periodic or quasi-periodic variability outside the dimming event. We ran a single-band Lomb −Scargle periodogram (LSP;Lomb 1976;Scargle 1982) using gatspy (VanderPlas & Ivezić 2015) with a twocomponent Fourier model. Across each photometric bandpass, we got inconsistent periods at maximum power, which is likely due to the difference in cadence between each filter and survey, despite the identification of periods exceeding 90 days in our LSP analysis, indicative of pulsating red giants in accordance with Kepler light curves (Ceillier et al. 2017). We calculated the significance of the peak through a bootstrap extrapolation and found that it fell below the 2σ threshold.

Light-curve Properties: Optical to IR
Given the remarkably long duration of the dimming event of Gaia17bpp, there are five photometric bandpasses that extensively covered the second half of the dimming event.
The bottom panel of Figure 3 shows how stitching together detections from multiple surveys and photometric bandpasses reveals almost the entirety of the dimming event except for the ingress. Neither WISE nor NEOWISE light curves showed any evidence of large-amplitude variability during both the occultation and dimming phases. Based on a 30-day median sliding window, we found that the light curve did not vary more than 0.2 mag across both the W1 and W2 magnitudes. The W1 and W2 light curves returned back to their median baseline brightness after the dimming event. Similarly, the Gaia G light curve did not exhibit any structure during the dimming event. Due to the sparse coverage of all presented photometric measurements, we recognize that there might exist variability on shorter timescales that are not captured.
Perhaps the more puzzling property of the Gaia17bpp system is the variation in colors across different epochs. In Figure 8 we show the color evolution of the dimming event, including the nondimming phase in three photometric bandpasses. In order to get the same epochs for the WISE and Gaia photometry, we binned the detection by a running median of 100 days. We first begin to note the constant color during the quiescent phase found in any epoch beyond 2019 January; this constant color profile is the inherent color of the primary star including Galactic foreground extinction. We do not observe any significant color variations within this phase. On the other hand, both the egress and bottom of the dimming event show substantial color variation. For example, in the top panel of Figure 8 we observe that during the minimum of the dimming event the (W1-W2) color was different by ∼0.2 mag compared to the color of the primary star outside the dimming event. We suggest that the shorter-wavelength W1 (3.4 μm) magnitude was brighter compared to the longer-wavelength W2 (4.7 μm) magnitude, indicating the excess shorter-wavelength flux with respect to the primary star (henceforth referred to as the blue bump). Similarly, the Gaia-WISE colors also suggest a similar evolution and a small blue bump at the minimum of the dimming event at different rates. The last two detections are also somewhat curious with a 200-day color 0.1 mag variation from blue to red; however, it is difficult to tell whether these are spurious detections or related to the dimming mechanism. Figure 5. SED fit (black synthetic PHOENIX spectrum) to the obtained broadband photometry outside the Gaia17bpp dimming event. We also include a simple blackbody fit to the photometry accounting for the effects of extinction at the line of sight. The bottom panel represents the residual levels from the PHOENIX model. We note that the SED fitting model does not extend to the W3 detection and W4 upper limit.
Given the five unique photometric filters we have that captured the egress of the dimming phase, we suspect that the shape of the light curves can possibly change from optical to infrared assuming that the obscuring process does not have an underlying gray spectrum. To test this hypothesis, we fit the overall properties of the light curve with a generalized Gaussian distribution (GGD): , 1 x 1 with light-curve amplitude (), mean (μ) time, standard deviation (α) that controls the width of the distribution, and the overallshape (β) of the GGD. The mathematical properties of the GGD are presented by Dytso et al. (2018); see also Cadirci et al. (2022). The power of the underlying GGD is the ability to control the flatness of the Gaussian peak since we can visually tell that the bottom of the dimming event was flat. To account for any unwanted correlated noise measurements, we simultaneously model the underlying correlated noise with a Gaussian process (GP) with George 7 (Ambikasaran et al. 2015) including the mean model parameters via Markov Chain Monte Carlo (MCMC). We use Emcee (Foreman-Mackey et al. 2013), which implements the Metropolis−Hasting algorithm used to constrain the posterior distribution. The correlated noise that might be present in the multiband light curves was captured by a Matérn-3/2 covariance function. We simultaneously fit the GP hyperparameters, including the underlying model.
Overall, the GGD model provided a good fit for the Gaia RP, Gaia G, W1, and W2 light curves as seen in Figure 10. Based on our GP analysis, we found clear deviations from the shape of the light curves across each bandpass. For example, in Figure 9 we show the posterior distributions for the model mean time, amplitude, standard deviation, and shape of the GGD profile for each light curve. According to the interquartile range of each distribution, the mean time, standard deviation, and amplitude show variations. Based on the wide distributions we obtained from the model mean epoch, all light curves had a minimum flux of around 2016 June (57200 MJD) within a margin of error of ∼400 days. We notice that the WISE light curves have a mean time different by ∼150 days compared to the optical Gaia light curves. It is noticeable how the variance of the α posterior distribution shifts from the optical 550 to 450 in the infrared, suggesting that the optical light curves have a slightly longer duration. Thus, we find that the light-curve profile of the Gaia17bpp dimming event is modestly asymmetric. The modest asymmetric dimming profile likely highlights a weak wavelength dependence that reflects the intrinsic properties of the obscuring object. We note that the narrow range of posterior distribution, particularly for the less complete Gaia G and RP light curves, is reflected in our assumed light-curve shape. We could not successfully place firm limits on the BP light curve since it suffers from a large scatter. In Figure 10 we display our GP GGD fits drawn from the posterior distribution for the RP, Gaia G, W1, and W2 light curves including their residuals.
Given the poor photometric coverage at the minimum of the dimming event, we do not have enough multiband photometry to perform any SED inference. We attempted to fit a simple blackbody model to the BP, G, RP, PS1-y, W1, and W2 detections and found that the temperature was consistent with the findings of the dimming phase at 3800 K. We only notice that the dimming detections are roughly equivalent to the nondimming detections scaled down by some factor.

Intrinsic Variability Scenarios
Month-to-year-long dimming events have been identified from rare hydrogen-deficient carbon-rich stars, R Coronae Borealis (RcB) type variables that eject large carbon-dust clouds. RcB stars are known for their erratic and asymmetric dimming profiles and are dominated by mostly helium, nitrogen, and carbon enrichment in their atmospheres (De Marco et al. 2002). We found no He I emission or P Cygni profile in our optical spectrum. RcB stars are also known to exhibit mid-IR excess due to warm circumstellar dust shells found around them (Tisserand 2012). No carbon absorption features or other spectroscopic RcB signatures were found in our optical spectrum. Considering the stellar SED of Gaia17bpp, we also do not find any evidence of mid-IR excess. We examined the WISE color-color-magnitude diagram of the primary star to see whether the colors were consistent with known RcB stars. We compared the (W1-W2) 0 and (W3-W4) 0 , with values −0.17 and 1.1, respectively, and found that the observed color is 0.5 mag away from where typical RcB stars have been found in the WISE catalog (Tisserand et al. 2020). Given the additional absence of any other erratic variability and the nonasymmetric light curve of Gaia17bpp, we find it unlikely that it is an RcB candidate.
Red supergiants (RSGs) are also known to exhibit largeamplitude dimming events that can last up to a few months to a year (Massey et al. 2007;Dupree et al. 2022). One particular case study is the great dimming event of Betelgeuse, which exhibited a deep 2-month-long dimming event. Initial interpretations noticed that the temperature of Betelgeuse had cooled down to 3600 K and suggested that surface convection effects caused the dimming event. Levesque & Massey (2020) discussed that a large and cool convection cell on Betelgeuse would have caused strong TiO bands with a substantially lower temperature; instead, the leading hypothesis is an episodic mass-loss event with large-grain circumstellar dust. Our current analysis does not favor an RSG classification for Gaia17bpp given its modest radius and low-mass estimates based on the SED. Similarly, the duration of the dimming event we observe is not compatible with the timescales currently reported in the literature on dimming RSG stars (Kiss et al. 2006). We find it unlikely that the Gaia17bpp dimming mechanism was caused by the ejection of a dust cloud around RSG stars.

Extrinsic Variability Scenarios
Our current available data do not provide evidence that Gaia17bpp is a compact X-ray binary or an energetic cataclysmic variable. We searched through known X-ray catalogs (ROSAT 2RXS, XMMDR10, XMMSL2, Chandra CSC2, swift2SXPS) and found no sources at the location of Gaia17bpp except for a 2σ upper limit from the Swift/XRT count rate at 0.36 counts s −1 . We derived an X-ray limiting luminosity of approximately 10 35 erg s −1 at the distance of the source. 8 Given the relatively shallow upper X-ray flux limit at this line of sight, it is possible that Gaia17bpp is an active X-ray source but has been missed owing to the lack of deeper limits. Other common telltale signs of accretion are usually seen in Hα emission from an ionized disk (Williams 1980). As seen in Figure 4, we do not see any narrow Hα or other emission lines of ongoing accretion. Given the lack of erratic photometric variability, no X-ray detections, and no evidence of narrow emission lines in the optical spectrum at the time, we do not believe that the dimming of Gaia17bpp results from the activity in an accretion disk.
Young stellar objects (YSOs) are systematically known to display deep and complex dimming events on timescales of months to years called dipper events (Cody & Hillenbrand 2018). In most cases, YSO disks are also luminous X-ray and radio sources owing to the strong magnetic fields and high temperatures from ongoing accretion (Feigelson & Montmerle 1999). Our archival data do not support a YSO scenario given the lack of infrared excess and the lack of bright nearby radio or X-ray sources. Since the typical timescales of YSO disks are a few megayears, we do not expect to find such a young companion near an evolved giant star. Finally, the sky position and estimated distance of Gaia17bpp are not near any known star-forming regions.

Occulting Disk Model
As discussed in Sections 5.1 and 5.2, there is no currently known intrinsic or extrinsic dimming mechanism outside the M giant that can cause stars to smoothly dim for prolonged periods. We will now focus on the most probable cause of the dimming event, which could be attributed to an occultation by a disk. We find that this explanation could account for the prolonged dimming event observed fromGaia17bpp.
The smooth and asymmetric dimming event of Gaia17bpp can be plausibly explained by the transit of an oblate disk similar to that of ò Aur analog systems. For example, Rappaport et al. (2019) showed that EPIC 204376071, a young M-type star, was obscured by an oblate tilted dusty disk. To compute the transiting models, they used a modified version of pyPplusS (Rein & Ofir 2019) that computes light curves for oblate spherical exoplanets that possess rings and can account for both uniform and limb-darkened scenarios that are performed using the Polygon-Segments algorithm. Rappaport et al. (2019) expanded this tool by adding a full opacity layer to the rings to simulate the transit of an oblate disk.
We attempted to apply this suite of models to the W1 light curve of Gaia17bpp since it contains detections from all phases of the event. One of the key modifications we made to the existing code was to also include in the fitted light curve the limb-darkening coefficient. Given the priors we obtained from our SED analysis, we only consider the c 2 coefficient of the quadratic limb-darkening equation since the first linear term is very small. 9 For the purposes of this study, we only fit a model of an oblate solid disk model with seven parameters: disk radius in radii of the primary star (R disk ), impact parameter (b), orbital inclination (i), the tilt angle of the disk (f), transverse Figure 8. Color evolution of Gaia17bpp in the infrared and infrared-to-optical photometry. We highlight in gray the primary region of the dimming event and in red the egress of the dimming event. In each panel, the underlying solid black line model was produced from a generalized Gaussian distribution described in Section 4. speed of the disk across the host star (v), time shift (dx), and the quadratic polynomial limb-darkening term (c 2 ). We explored a variety of different model parameters that could produce a 7 yr long dimming event. We found that the models that could produce such long events needed to have low transverse velocities that likely indicate the large semimajor axis between the occulter and primary star. We proceeded to perform an MCMC implementation using a standard Gaussian likelihood function. In the limitation of poor constraints on the geometry of the system, we assumed broad uniform priors for all parameters. In Figure 11 we show the model fit against the data. While the disk light-curve model performs overall a reasonably good fit to the data, we do find some weakly correlated noise within the residuals (bottom panel of Figure 11). Considering the limited constraints at our disposal and the prior orbital configuration of this system, it remains very challenging to constrain based on the current data. Nonetheless, we found that the semimajor axis of the disk is roughly 1.4 au, including the angle between the semimajor axis and the direction of motion across the sky to be inclined by ∼178°. It is thus possible that we are seeing the disk obscuring the star edge-on since we also find the impact parameter to be small (b ∼ 0). The marginalized posterior distribution also suggests that the transverse velocity is low, approximately at 0.005 km s −1 . In Table 4, we summarize the marginalized posterior distribution from our analysis. We attempted to repeat the same model with the exception of introducing an additional ring to the oblate occulter with some tunable opacity term. We found that such a two-layer oblate occulter was not capable of fitting the overall shape of the light curve owing to the steep transition between a solid disk and another additional optically    Note. For the posterior distribution, we report the median and the 1σ range. thin layer. Based on the performance of our single oblate transiting model, we have reasonable evidence to suggest that the occulting object of Gaia17bpp must have been an extended object such as an oblate disk. However, in order to establish more robust constraints on the geometry of the disk, additional multiband light-curve modeling will be needed. We have reasonable data to challenge the unlikely scenario of a circumbinary precessing disk such as the case of KH 15Dlike analogs (Windemuth & Herbst 2014;Zhu et al. 2022). For example, almost all known KH 15D-like analogs have been found to have characteristic shallower infrared light curves, including clear excess in infrared. Additionally, given the collected photometry, we were unable to find any short-period signals indicating the presence of the companion. One similarity between Gaia17bpp and KH 15D-like analogs is the blue-bump feature in the infrared light curves. Based on the archival data alone, we are unable to identify the mechanism behind the blue bump observed during the minimum of the Gaia17bpp dimming event. One possible scenario is the presence of another hot source contributing to the observed color at the bottom of the dimming event. A second scenario is the effect of Mie forward scattering due to the dust grain size that is comparable to the wavelength (Silvia & Agol 2008). It is thus indicated that the characteristic grain size of the Gaia17bpp occulting disk is likely larger than the average foreground interstellar medium (Arulanantham et al. 2016). Compared to other possible candidates of the same nature, for example, ò Aur (Carroll et al. 1991, see Figure 5 therein), which has clear amplitude variations during the maximum eclipse, Gaia17bpp shows overall smaller variability that might shed light on the geometry and optical depth of the occulter. Gaia17bpp and ò Aur also show similarities in asymmetries in their light-curve properties throughout different bandpasses, with ò Aur having an ingress dimming and somewhat steeper egress brightening phase (Stencel 2012). Given our findings on the modestly asymmetric dimming event across the multiband light curves of Gaia17bpp, we hypothesize that a weak wavelength dependence might be connected to the overall properties of the occulting disk (Parthasarathy & Frueh 1986). It has been found that debris disks may lack infrared emission owing to the nonradial distribution of mass, compared to typical protoplanetary disks (Hughes et al. 2018). Stencel et al. (2011) attempted to find evidence for an infrared excess or silicate features in ò Aur; however, the SED was found to be smooth and lacked classic dust features. It is suggested that large particles in a debris disk in ò Aur could account for the observed SED. Both cases of ò Aur and TYC 2505-672-1 have evidence to suggest the scenario of a debris disk around the secondary star. Thus, this would also be in line with a lack of infrared excess that we notice with the Gaia17bpp system. Given the limitations of the archival data presented here, future far-IR studies with more sensitivity should investigate whether the presence of a cold optically thick disk exists beyond the wavelength ranges explored in this work.
Given Gaia's superb microarcsecond resolution, if a luminous companion existed, it would likely be detected by Gaia. For instance, assuming a mass of 1.5 M e and a minimum orbital period of 66 yr, we estimate the minimum angular separation to be 516 μas, which is within the resolving capabilities of Gaia. We searched a 0.1 deg 2 overlapping area with Gaia17bpp through both Gaia DR2 and DR3 catalogs to look for any detected sources that might be associated with this system. We identified two faint sources (Gaia DR3 1636148068921376768 with G mag = 21.1, and Gaia DR3 1636148068921376768 with G mag = 20.5) within this cone search; however, they lacked astrometric solutions or had large parallax measurements. While we are unable to confirm the association of these sources with Gaia17bpp, future studies should examine more closely their possible association. On the other hand, the nondetection of the secondary star was not a surprise. First, we are limited to shallow NUV detection owing to the high line-of-sight extinction and distance of Gaia17bpp. Second, if the secondary is surrounded by an extended disk, then it is possible that the companion star is hiding within the disk, such as the case of ò Aur identified B5V stellar companion (Stencel 2012). We constructed a simple toy model of a blackbody mixture model that incorporates the blackbody emission flux from the primary and a hypothetical secondary at the same distance, including the effects of Galactic attenuation. In Figure 12 we vary the temperature and radius of the hypothetical secondary star and integrate the blackbody mixture model to estimate the approximate SDSS u AB magnitude. We include in Figure 12 the derived SDSS u ∼ 22 mag from our images as a dashed line. Based on our limiting magnitude, we cannot exclude the presence of a main-sequence, WD, or neutron star companion based on the observations alone. Further deep observations in the NUV and far-UV will be required.
In Figure 13 we compare the relative photometric depth of Gaia17bpp with other giant dimming stars obscured by occulting disks, sorted from the shortest-to longest-duration dimming events. It is evident that the Gaia17bpp system has by far the longest and deepest dimming event found among these candidates. It is possible that the variation in light-curve properties we see here is due to the geometric configurations of such disk companions. Based on the archival photometry alone, it is unclear whether Gaia17bpp is periodic. Based on the archival detections from DASCH and other surveys, we find it likely that no other major dimming event has occurred. Given the detections we have made from DASCH and POSS, we can assume a lower limit to the orbital period of at least 66 yr. This would be in agreement with our occulting disk model, which would suggest a very long orbital period given the very low transverse velocity. Figure 12. Grid of blackbody mixture model including the radius and effective temperatures of the secondary star color-coded by the integrated SDSS u AB magnitude. The dashed line represents our limiting SDSS u-band magnitude. We assume that the primary star has a fixed radius of 58 R e , an effective temperature of 3850 K, and a median distance of 8965 pc.

Conclusion
In conclusion, we report the serendipitous discovery of Gaia17bpp/2MASS J19372316+1759029, an anomalous binary star that exhibited a single, ∼7 yr long, 4.5 mag deep dimming event. Using a collated data set of multiband light curves across several surveys and wavelength regimes, we find that this unique system closely resembles the variability seen from evolved giant stars transited by companion stars with opaque disks. We constrain the primary star of Gaia17bpp through a Bayesian amalgamation of stellar SED models. We conclude that the primary star is likely a 58 R e M0III giant star favoring a low-mass primary star that is consistent with the radius and mass profile of an ordinary giant M giant star. The long and complex nature of the dimming event is not fully understood. We find evidence that the WISE IR light curves are shallower, asymmetric, and slightly different shapes from their optical counterparts. The color evolution of the dimming event is also perplexing in both optical and IR colors, which is not fully understood. We suspect that the main culprit of the dimming event is linked to an emerging population of rare binary systems with a companion enshrouded in a debris disk, such as the case of ò Aur. Finally, we performed a simple tilteddisk eclipse model of the W1 infrared light curve. We found that the disk would need to be inclined by 178°and slow moving with an approximate radius of 1.4 au. Based on the currently available data, we cannot identify a secondary companion star.
Slow and photometrically deep dimming stellar systems will become even more relevant in the near future. The forthcoming Vera C. Rubin Observatory Legacy Survey of Space and Time (LSST; Ivezić et al. 2019) promises a survey of long duration and unprecedented photometric depth. LSST will contain both a long-time baseline and unprecedented photometric depth in ugrizy that will lead to the discovery of many more such extraordinary eclipsing systems.