Breaking the Habit: The Peculiar 2016 Eruption of the Unique Recurrent Nova M31N 2008-12a

Like the 2014 and 2015 eruptions before it (DHS15, DHB16), the 2016 eruption of M31N 2008-12a was observed by a large number of ground-based telescopes operating in the visible regime. Unfortunately, due to poor weather conditions at many of the planned facilities, observations of the 2016 eruption were much sparser than in recent years.

A major achievement for the 2016 eruption campaign was the addition of extensive observations from the American Association of Variable Star Observers (AAVSO⁵¹ ), along with the continued support of the Variable Star Observers League in Japan (VSOLJ⁵² ; see Section 3.1 and Appendix A). Observations were also obtained from the Mount Laguna Observatory (MLO) 1.0 m telescope in California, the Ondřejov Observatory 0.65 m telescope in the Czech Republic, the Danish 1.54 m telescope at La Silla in Chile, the fully robotic 2 m Liverpool Telescope (LT; Steele et al. 2004) in La Palma, the 2.54 m Isaac Newton Telescope (INT) at La Palma, the Palomar $48^{\prime\prime}$ telescope in California, the 0.6 m and 1 m telescopes operated by members of the Embry Riddle Aeronautical University (ERAU) in Florida, the $2\times 8.4$ m (11.8 m eq.) Large Binocular Telescope (LBT) on Mount Graham, Arizona, the 2 m Himalayan Chandra Telescope (HCT) located at Indian Astronomical Observatory (IAO), Hanle, India, and the 2.4 m Hubble Space Telescope.

2.1.1. Hubble Space Telescope Photometry

2.1.2. Ground-based Photometry

The spectroscopic confirmation of the 2016 eruption of M31N 2008-12a was announced by Darnley et al. (2016c), with additional spectroscopic follow-up reported in Pavana & Anupama (2016). A summary of all optical spectra of the 2016 eruption of M31N 2008-12a is shown in Table 3; all spectra are reproduced in Figure 16.

Table 3.  Summary of the Optical Spectra of the 2016 Eruption of M31N 2008-12a

Date (UT)	${\rm{\Delta }}t$	Instrument	Exposure
2017 Dec	(days)	and Telescope	Time (s)
12.86	0.54 ± 0.01	ALFOSC/NOT	1 × 1500
12.93	0.61 ± 0.06	SPRAT/LT	6 × 900
13.14	0.82 ± 0.11	DIS/ARC
13.61	1.29 ± 0.02	HFOSC/HCT	1 × 3600
13.98	1.66 ± 0.07	SPRAT/LT	6 × 900
14.12	1.80 ± 0.08	DIS/ARC
14.19	1.87 ± 0.02	LRS2-B/HET	3 × 600
14.55	2.23 ± 0.02	HFOSC/HCT	1 × 2700
14.90	2.58 ± 0.05	SPRAT/LT	6 × 900
15.91	3.59 ± 0.02	SPRAT/LT	3 × 900
16.85	4.53 ± 0.02	SPRAT/LT	3 × 900
18.15	5.83 ± 0.05	DIS/ARC

Note. The time since eruption assumes an eruption date of 2016 December 12.32 UT. The error bars do not include the systematic error in this eruption date, but represent the total exposure time/time between combined exposures of a given epoch.

Download table as:  ASCIITypeset image

We obtained several spectra of the 2016 eruption with SPRAT (Piascik et al. 2014), the low-resolution, high-throughput spectrograph on the LT. SPRAT covers the wavelength range of 4000–8000 Å and uses a $1\buildrel{\prime\prime}\over{.} 8$ slit, giving a resolution of ~18 Å. We obtained our spectra using the blue-optimized mode. The data were reduced using a combination of the LT SPRAT reduction pipeline and standard routines in IRAF⁵⁴ (Tody 1993). The spectra were calibrated using previous observations of the standard star G191-B2B against data from Oke (1990) obtained via ESO. Conditions on La Palma were poor during the time frame the nova was accessible with SPRAT during the 2016 eruption, so the absolute flux levels are possibly unreliable.

We obtained an early spectrum of the nova, 0.54 days after eruption, using the Andalucía Faint Object Spectrograph and Camera (ALFOSC) on the 2.5 m Nordic Optical Telescope (NOT) at the Roque de los Muchachos Observatory on La Palma. Grism #7 and a slit width of $1\buildrel{\prime\prime}\over{.} 3$ yielded a spectral resolution of 8.5 Å at the center of the useful wavelength range 4000–7070 Å ($R\sim 650$). The 1500 s spectrum was imaged on the 2048 × 2048 pixel CCD #14 with 2 × 2 binning. We performed the observation under poor seeing conditions ($\sim 2\buildrel{\prime\prime}\over{.} 5$). We reduced the raw images using standard IRAF procedures, and then did an optical extraction of the target spectrum with starlink/pamela (Marsh 1989). The pixel-to-wavelength solution was computed by comparison with 25 emission lines of the spectrum of a HeNe arc lamp. We used a fourth-order polynomial that provided residuals with an rms more than 10 times smaller than the spectral dispersion.

In addition, 1.87 days after eruption, we obtained a spectrum of M31N 2008-12a using the blue channel of the 10 m Hobby Eberly Telescope's (HET's) new integral-field Low Resolution Spectrograph (LRS2-B; Chonis et al. 2014, 2016). This dual-beam instrument uses 280 fibers and a lenslet array to produce spectra with a resolution of $R\sim 1910$ between the wavelengths 3700 and 4700 Å, and $R\sim 1140$ between 4600 and 7000 Å over a $12^{\prime\prime} \times 6^{\prime\prime}$ region of sky. The seeing for our observations was relatively poor ($1\buildrel{\prime\prime}\over{.} 8$), and the total exposure time was 30 minutes, split into three 10 minute exposures.

Reduction of the LRS2-B data was accomplished using Panacea,⁵⁵ a general-purpose IFU reduction package built for HET. After performing the initial CCD reductions (overscan removal and bias subtraction), we derived the wavelength solution, trace model, and spatial profile of each fiber using data from twilight-sky exposures taken at the beginning of the night. From these models, we extracted each fiber's spectrum and rectified the wavelength to a common grid. Finally, at each wavelength in the grid, we fit a second-order polynomial to the M31's background starlight and subtracted that from the Gaussian-shaped point source assumed for the nova.

Two epochs of spectra were obtained using the Himalayan Faint Object Spectrograph and Camera (HFOSC) mounted on the 2 m HCT located at IAO, Hanle, India. HFOSC is equipped with a 2k × 4k E2V CCD with pixel size of 15 × 15 μm. Spectra were obtained in the wavelength range 3800–8000 Å on 2016 December 13.61 and 14.55 UT. The spectroscopic data were bias subtracted and flat-field corrected and extracted using the optimal extraction method. An FeAr arc lamp spectrum was used for wavelength calibration. The spectrophotometric standard star Feige 34 was used to obtain the instrumental response for flux calibration.

Three spectra were obtained with the 3.5 m Astrophysical Research Consortium (ARC) telescope at the Apache Point Observatory, during the first half of the night on 2016 December 12, 13, and 17 (UT December 13, 14, and 18). We observed with the Dual Imaging Spectrograph (DIS): a medium-dispersion long-slit spectrograph with separate collimators for the red and blue parts of the spectrum and two 2048 × 1028 E2V CCD cameras, with the transition wavelength around 5350 Å. For the blue branch, a 400 line mm⁻¹ grating was used, while the red branch was equipped with a 300 line mm⁻¹ grating. The nominal dispersions were 1.83 and 2.31 Å pixel⁻¹, respectively, with central wavelengths at 4500 and 7500 Å. The wavelength regions actually used were 3500–5400 Å and 5300–9900 Å for blue and red, respectively. A $1\buildrel{\prime\prime}\over{.} 5$ slit was employed. Exposure times were 2700 s. At least three exposures were obtained per night. Each on-target series of exposures was followed by a comparison lamp exposure (HeNeAr) for wavelength calibration. A spectrum of the spectrophotometric flux standard (BD +28 4211) was also acquired during each night, along with bias and flat-field calibration exposures. The spectra were reduced using Python scripts to perform standard flat-field and bias corrections to the 2D spectral images. Extraction traces and sky regions were then defined interactively on the standard star and object spectral images. Wavelength calibration was determined using lines identified on the extracted HeNeAr spectra. We then determined the solution by fitting a third-order polynomial to these measured wavelengths. Flux calibration was determined by measuring the ratio of the star fluxes to the known fluxes as a function of wavelength. We performed these calibrations independently for the red and blue spectra, so that the clear agreement in the overlapping regions of the wavelength ranges confirms that our calibration and reduction procedure was successful.

A Neil Gehrels Swift  Observatory (Gehrels et al. 2004) target of opportunity (ToO) request was submitted immediately after confirming the eruption, and the satellite began observing the nova on 2016 December 12.65 UT (cf., Henze et al. 2016d), only four hours after the optical discovery. All Swift  observations are summarized in Tables 4 and 5. The Swift  target ID of M31N 2008-12a  is always 32613. Because of the low-Earth orbit of the satellite, a Swift  observation is normally split into several snapshots, which we list separately in Table 11.

Table 4.  Swift  Observations of M31N 2008-12a for the 2016 Eruption

ObsID	Expa	Dateb	MJDb	${\rm{\Delta }}t$ c	uvw2d	XRT Ratee
	(ks)	(UT)	(day)	(day)	(mag)	(${10}^{-2}$ ct s−1)
00032613183	3.97	2016 Dec 12.65	57734.65	0.33	16.7 ± 0.1	<0.3
00032613184	4.13	2016 Dec 13.19	57735.19	0.87	17.3 ± 0.1	<0.2
00032613185	3.70	2016 Dec 14.25	57736.26	1.94	17.9 ± 0.1	<0.3
00032613186	3.23	2016 Dec 15.65	57737.65	3.33	18.6 ± 0.1	<0.4
00032613188	1.10	2016 Dec 16.38	57738.38	4.06	18.7 ± 0.1	<0.7
00032613189	3.86	2016 Dec 18.10	57740.10	5.78	19.3 ± 0.1	0.6 ± 0.1
00032613190	4.03	2016 Dec 19.49	57741.50	7.18	20.0 ± 0.2	0.4 ± 0.1
00032613191	2.02	2016 Dec 20.88	57742.89	8.57	20.6 ± 0.3	1.9 ± 0.3
00032613192	3.95	2016 Dec 21.49	57743.49	9.17	20.9 ± 0.3	1.5 ± 0.2
00032613193	2.53	2016 Dec 22.68	57744.69	10.37	20.4 ± 0.2	1.7 ± 0.3
00032613194	2.95	2016 Dec 23.67	57745.68	11.36	20.8 ± 0.3	1.4 ± 0.2
00032613195	2.90	2016 Dec 24.00	57746.01	11.69	20.5 ± 0.2	0.7 ± 0.2
00032613196	2.73	2016 Dec 25.00	57747.01	12.69	>21.1	0.6 ± 0.2
00032613197	2.71	2016 Dec 26.20	57748.20	13.88	>21.1	0.3 ± 0.2
00032613198	2.84	2016 Dec 27.72	57749.73	15.41	>21.1	<0.5
00032613199	3.23	2016 Dec 28.19	57750.19	15.87	>21.2	<0.4
00032613200	2.65	2016 Dec 29.45	57751.46	17.14	>21.1	<0.5
00032613201	3.05	2016 Dec 30.05	57752.05	17.73	>20.9	<0.4
00032613202	2.88	2016 Dec 31.58	57753.58	19.26	>21.1	<0.3

Notes.

^aExposure time includes dead-time corrections. ^bObservation start date. ^cTime in days after the eruption date on 2016 December 12.32 UT (MJD 57734.32) ^dThe Swift  UVOT uvw2 filter has a central wavelength of 1930 Å with an FWHM of about 660 Å. ^eCount rates are measured in the 0.3–1.5 keV range.

Download table as:  ASCIITypeset image

Table 5.  Stacked Swift  UVOT Observations and Photometry as Plotted in Figure 5

ObsIDsa	Expb	Datec	MJDc	${\rm{\Delta }}t$ c	Lengthd	uvw2
	(ks)	(UT)	(day)	(day)	(day)	(mag)
00032613196/198	8.3	2016 Dec 26.37	57748.37	14.05	2.72	21.7 ± 0.4
00032613199/200	5.9	2016 Dec 28.83	57750.83	16.51	1.27	<21.5

Notes.

^aStart/end observation for each stack (cf., Table 4). ^bSummed-up exposure. ^cTime between the eruption date (MJD 57734.32; cf. Section 3.1) and the stack midpoint. ^dTime in days from the first observation of the stack to the last one.

Download table as:  ASCIITypeset image

In addition, we triggered a 100 ks XMM-Newton  (Jansen et al. 2001) ToO that was originally aimed at obtaining a high-resolution X-ray spectrum of the SSS variability phase. Due to the inconvenient eruption date, 14 days before the XMM-Newton  window opened, and the surprisingly fast light curve evolution, discussed in detail below, only low-resolution spectra and light curves could be obtained. The XMM-Newton  object ID is 078400. The ToO was split into two observations, which are summarized in Table 6. Since 2008, no eruption of M31N 2008-12a  had occurred within one of the relatively narrow XMM-Newton  visibility windows from late December to mid-February, and from July to mid-August (cf., Table 1).

Table 6.  XMM-Newton  Observations of M31N 2008-12a in 2016

ObsID	Expa	GTIb	MJDc	${\rm{\Delta }}t$ d	uvw1e	EPIC Rate	Equivalent XRT Ratef
	(ks)	(ks)	(UT)	(d)	(mag)	(${10}^{-2}$ ct s−1)	(${10}^{-4}$ ct s−1)
0784000101	33.5	16.1	57748.533	14.21	${21.6}_{-0.2}^{+0.3}$	1.9 ± 0.2	7.3 ± 0.6
0784000201	63.0	40.0	57750.117	15.80	21.6 ± 0.2	1.0 ± 0.1	3.3 ± 0.2

Notes.

^aDead-time corrected exposure time for XMM-Newton  EPIC pn prior to GTI filtering for high background. ^bExposure time for XMM-Newton  EPIC pn after GTI filtering for high background. ^cStart date of the observation. ^dTime in days after the eruption of nova M31N 2008-12a  in the optical on 2016 December 12.32 UT (MJD = 57734.32; see Itagaki et al. 2016). ^eThe OM filter was uvw1 (central wavelength 2910 Å with a width of 830 Å.) ^fTheoretical Swift  XRT count rate (0.3–10.0 keV) extrapolated based on the 0.2–1.0 keV EPIC pn count rates in the previous column, and assuming the best-fit blackbody spectrum and foreground absorption.

Download table as:  ASCIITypeset image

The Swift  UV/optical telescope (UVOT; Roming et al. 2005) magnitudes were obtained via the HEASoft (v6.18) tool uvotsource, based on aperture photometry of carefully selected source and background regions. We stacked individual images using uvotimsum. In contrast to previous years, our 2016 coverage exclusively used the uvw2 filter, which has a central wavelength of 1930 Å. The photometric calibration assumes the UVOT photometric (Vega) system (Poole et al. 2008; Breeveld et al. 2011) and has not been corrected for extinction.

In the case of the Swift  X-ray telescope (XRT; Burrows et al. 2005) data, we used the online software⁵⁶ of Evans et al. (2009) to extract count rates and upper limits for each observation and snapshot, respectively. Following the recommendation for SSSs, we extracted only grade-zero events. The online software uses the Bayesian formalism of Kraft et al. (1991) to estimate upper limits for low numbers of counts. All XRT observations were taken in the photon counting (PC) mode.

The XMM-Newton  X-ray data were obtained with the thin filter for the pn and MOS detectors of the European Photon Imaging Camera (EPIC; Strüder et al. 2001; Turner et al. 2001). They were processed with XMM-SAS (v15.0.0) starting from the observation data files and using the most recent current calibration files (CCF). We used evselect to extract spectral counts and light curves from source and background regions that were defined by eye on the event files from the individual detectors. We filtered the event list by extracting a background light curve in the 0.2–0.7 keV range (optimized after extracting the first spectra; see Section 4.2) and removing the episodes of flaring activity.

In addition, we obtained UV data using the XMM-Newton  optical/UV monitor telescope (OM; Mason et al. 2001). All OM exposures were taken with the uvw1 filter, which has a slightly different but comparable throughput to the Swift  UVOT filter of the same name (cf., Roming et al. 2005). The central wavelength of the OM uvw1 filter is 2910 Å with a width of 830 Å (cf., UVOT uvw1: central wavelength 2600 Å, width 693 Å; see Poole et al. 2008). We estimated the magnitude of M31N 2008-12a  in both observations via carefully selected source and background regions, which were based on the Swift  UVOT apertures. Our estimates include (small) coincidence corrections and a PSF curve-of-growth correction. The latter became necessary because the size of the source region needed to be restricted to avoid contamination by neighboring sources. The count rate and uncertainties were converted to magnitudes using the CCF zero points.

As in previous papers on this object (HND14, HND15, DHB16), the X-ray spectral fitting was performed in XSPEC (v12.8.2; Arnaud 1996) using the Tübingen–Boulder ISM absorption model (TBabs in XSPEC) and the photoelectric absorption cross-sections from Balucinska-Church & McCammon (1992). We assumed the ISM abundances from Wilms et al. (2000) and applied Poisson likelihood ratio statistics (Cash 1979).

With a nova that evolves as rapidly as M31N 2008-12a, early detection of each eruption is crucial. Following the successful eruption detection campaigns for the 2014 and 2015 outbursts, in 2016 we grew our large, multifacility monitoring campaign into a global collaboration. The professional telescopes at the LT, Las Cumbres (LCO; Brown et al. 2013, the 2 m at Haleakala, Hawai'i and the 1 m at McDonald, Texas), and Ondřejov Observatory were joined by a network of highly motivated and experienced amateur observers in Canada, China, Finland, Germany, Italy, Japan, the United Kingdom, and the United States. A large part of their effort was coordinated through the AAVSO and VSOLJ, respectively (see Appendix A for details). The persistence of the amateur observers in our team, during six suspenseful months of monitoring, allowed us to discover the eruption at an earlier stage than in previous years.

The 2016 eruption of M31N 2008-12a was first detected on 2016 December 12.4874 (UT) by the 0.5 m f/6 telescope at the Itagaki Astronomical Observatory in Japan at an unfiltered magnitude of 18.2 (Itagaki 2016). The previous nondetection took place at the LCO 1 m (McDonald) just 0.337 days earlier, providing an upper limit of $r^{\prime} \gt 19.1$. A deeper upper limit of $u^{\prime} \gt 22.2$ was provided by the LT and its automated real-time alert system (see Darnley et al. 2007) 0.584 days predetection. The 2016 eruption was spectroscopically confirmed almost simultaneously by the NOT and LT, 0.37 and 0.39 days postdetection, respectively (Darnley et al. 2016c).

All subsequent analysis assumes that the 2016 eruption of the nova M31N 2008-12a  (${\rm{\Delta }}t=0$) occurred on 2016 December 12.32 UT ($\mathrm{MJD}=57734.32$). This date is defined as the midpoint between the last upper limit (2016 December 12.15 UT; LCO) and the discovery observation (2016 December 12.49 UT; Itagaki observatory), as first reported by Itagaki et al. (2016). The corresponding uncertainty on the eruption date is ±0.17 days. The corresponding dates of the 2013, 2014, and 2015 eruptions, to which we will compare our new results, are listed in Table 1.

The HST photometry serendipitously obtained over the five-day pre-eruption period is shown in Figure 1. The Hα photometry is shown by the black points and the narrowband continuum by the red. Clear variability is seen during this pre-eruption phase. As this variability appears in both Hα and the continuum, it is possible that it is continuum driven. The system has a clear Hα excess immediately before eruption, but the Hα excess appears to diminish as the continuum rises. Following the discussion presented in DHG17P, it is possible that such Hα emission arises from the M31N 2008-12a accretion disk, which may be generating a significant disk wind.

Zoom In Zoom Out Reset image size

The continuum flux during this period is broadly consistent with the quiescent luminosity of the system (see DHG17P). Therefore, it is unclear whether this behavior is a genuine pre-eruption phenomenon or related to variability at quiescence with a characteristic timescale of up to a few days, with possible causes being accretion-disk flickering, or even orbital modulation. Through constraining the mass donor, DHG17P indicated that the orbital period for the M31N 2008-12a binary should be $\gtrsim 5$ days. Such a variation, as shown in Figure 1, would not be inconsistent with that constraint.

Following the 2015 eruption, DHB16 noted that the 2013, 2014, and 2015 eruption light curves, spanning from the I band to the near-UV (redder passbands only have data from 2015), were remarkably similar; see red data points in Figure 2. Based on those observations, DHB16 defined four phases of the light curve: the final rise (Days 0–1) is a regime sparsely populated with data due to the rapid increase to maximum light; the initial decline (Days 1–4), where a exponential decline in flux (linear in magnitude) is observed from the NUV to the near-infrared (see, in particular, the red data points in Figure 3); the plateau (Days 4–8), a relatively flat, but jittery, region of the light curve which is coincident in time with the SSS onset; and the final decline (Days > 8), where a power-law (in flux) decline may be present.

Zoom In Zoom Out Reset image size

Zoom In Zoom Out Reset image size

The combined 2013–2015 light curves defined these four phases; the individual light curves from each of those eruptions were also consistent with those patterns (see Figures 2 and 3). A time-resolved SED of the well-covered 2015 eruption was presented by DHB16. Unfortunately, due to severe weather constraints, our 2016 campaign did not obtain sufficient simultaneous multifilter data to compare the SED evolution. However, we find that the 2015 and 2016 light curves are largely consistent (Figure 2), except for the surprising features we will present in the following text.

First, we look at the initial decline phase for the 2016 eruption. We examine this region of the light curve first as, in previous eruptions, it has shown the simplest evolution—a linear decline—which was used by DHB16 to tie together the epochs of the 2013, 2014, and 2015 eruptions. But, due to the poor conditions at many of the planned sites, the data here are admittedly sparse, but are generally consistent with the linear behavior seen in the past three eruptions. There may however, be evidence for a deviation, approximately one magnitude upward, toward the end of this phase in the u'- and r'-band data at $t\gtrsim 3.6$ days post-eruption.

However, the largest deviation from the 2013 to 2015 behavior occurs during the final rise phase, between $0\leqslant t\leqslant 1$ days. There appears to be a short-lived, "cuspy" feature in the light curves seen through all filters (except in the B band where there was limited coverage) and unfiltered observations (see Figures 2–4, which progressively focus on the "cusp"). The variation between the peak luminosity of the 2013–2015 eruptions and the 2016 eruption is shown in Table 7; in all useful bands the deviation was significant. The average (across all bands) increase in maximum magnitude was 0.64 mag, or almost twice as luminous as the 2013–2015 eruptions at its peak. Notably, this overluminous peak occurred much earlier than the 2013–2015 peaks. The mean time of the peaks in 2013–2015 was $t\simeq 1.0$ day (across the u', B, R, r', and I filters), whereas the bright cusp in 2016 occurred at $t\simeq 0.65$ days.

Zoom In Zoom Out Reset image size

The INT and ERAU obtained a series of fast photometry of the 2016 eruption through g'-, i'- (ERAU only), and r'-band filters during the final rise phase. Figure 4 (left) compares this photometry with the 2013–2015 r'-band eruption photometry. This figure clearly illustrates not only the short-lived, bright, optical "cusp," but also its highly variable nature over a short timescale, with variation of up to 0.4 mag occurring over just 90 minutes. The $(g^{\prime} -r^{\prime} )$ color during this period is consistent with the cusp light curve being achromatic. We derive ${(g^{\prime} -r^{\prime} )}_{0}=0.15\pm 0.03$ for the cusp period, which is roughly consistent with the M31N 2008-12a color during the peak of the 2013–2015 eruptions (DHB16).

The 2013–2015 eruptions exhibited a very smooth light curve evolution from, essentially, t = 0 until $t\simeq 4$ days (see in particular the red r'-band light curve in Figure 3). As well as never being seen before, the bright cusp appears to break this smooth evolution. The 2016 eruption does not just appear more luminous than the observations of 2013–2015—there is evidence of a fundamental change, possibly in the emission mechanism, obscuration, or within the lines.

There are sparse data covering both the plateau and final decline phases. The R-band data from 2016 covers the entire plateau phase and is broadly consistent with the slow-jittery decline seen during this phase in the 2013–2015 eruptions. The u'- and r'-band data show a departure from the linear early decline around day 3.6; this could indicate an early entry into the plateau, i.e., a different behavior in 2016, or simply that the variation seen during the plateau always begins slightly earlier than the assumed four-day phase transition.

In essence, the 2016 light curves of M31N 2008-12a show a never before seen (but see Section 5.2.3), short-lived, bright cusp at all wavelengths during the final rise phase. There is no further strong evidence of any deviation from previous eruptions—however, we again note the sparsity of later-time data. Possible explanations for the early bright light curve cusp are discussed in Sections 5.2.1 and 5.3, and Section 5.2.3 re-examines earlier eruptions for possible indications of similar features.

During the 2015 eruption, we obtained a detailed Swift  UVOT light curve through the uvw1 filter (DHB16). For the 2016 eruption, our aim was to measure the uvw2 filter magnitudes instead to accumulate additional information on the broadband SED evolution. With a central wavelength of 1930 Å, the uvw2 band is the "bluest" UVOT filter (the uvw1 central wavelength is 2600 Å). Therefore, the spectral lines, for instance the prominent Mg ii (2800 Å) resonance doublet, affects the uvw1 range more than the uvw2 magnitudes (see DHG17S for details). Due to the peculiar properties of the 2016 eruption, a direct comparison between both light curves is now more complex than initially expected.

In Figure 5, we show the 2016 uvw2 light curve compared to the 2015 uvw1 (plus a few uvm2) measurements (DHB16) as well as a few uvw2 magnitudes from the 2014 eruption (HND15, DHS15). The 2016 values are based on individual Swift  snapshots (see Table 11) except for the last two data points where we used stacked images (see Table 5). Similarly to the uvw1 light curve in 2015, the uvw2 brightness initially declined linearly with ${t}_{2}=2.8\pm 0.2$ days. This is comparable to the 2015 uvw1 value of ${t}_{2}=2.6\pm 0.2$ days.

Zoom In Zoom Out Reset image size

From day three onward, the decline slowed down and became less monotonic. Viewed on its own, the UV light curve from this point onward would be consistent with a power-law decline (in flux) with an index of −1.5 ± 0.2. However, in light of the well-covered 2015 eruption, the 2016 light curve would also be consistent with the presence of three plateaus between (approximately) days 3–5, 6–8, and 9–12, and with relatively sharp drops of about 1 mag connecting those. Around day 12, when the X-ray flux started to drop (cf., Figure 6), there might even have been a brief rebrightening in the UV before it declined rapidly. The UV source had disappeared by day 16, which is noticeably earlier than in 2015 (in the uvw1 filter). DHG17P presented evidence that the UV–optical flux is dominated by the surviving accretion disk from at least day 13 onward. Therefore, a lower UV luminosity at this stage would imply a lower disk mass accretion rate. It is noteworthy that during the times when the 2014 and 2016 uvw2 measurements overlap, they appear to be consistent.

Zoom In Zoom Out Reset image size

The XMM-Newton  OM uvw1 magnitudes are given in Table 6 and included in Figure 5. The two OM measurements appear to be consistently fainter than the Swift  UVOT uvw1 data at similar times during the 2015 eruption (cf., DHB16). However, the uncertainties are large and the filter response curves (and instruments) are not perfectly identical. Therefore, we do not consider this apparent difference to have any physical importance. In addition, there is a hint of variability in the uvw1 flux during the first XMM-Newton  observation. Of the seven individual OM exposures, the first five can be combined to a uvw1 = ${21.3}_{-0.2}^{+0.3}$ mag, whereas the last two give a 2σ upper limit of uvw1 $\gt \,21.5$ mag. The potential drop in UV flux corresponds to the drop in X-ray flux after the peak in Figure 8. Also here the significance of this fluctuation is low, and we only mention it for completeness, in case similar effects will be observed in future eruptions.

Table 8.  FWHM Velocity Measurements of the Hα Profile During the 2016 Eruption

${\rm{\Delta }}t$ (days)	Hα FWHM (km s−1)	Instrument
0.54 ± 0.01	4540 ± 300	ALFOSC
0.61 ± 0.06	3880 ± 220	SPRAT
0.82 ± 0.11	3260 ± 130	DIS
1.29 ± 0.02	3010 ± 90	HFOSC
1.66 ± 0.07	3070 ± 120	SPRAT
1.80 ± 0.08	2910 ± 80	DIS
1.87 ± 0.02	2690 ± 60	LRS2-B
2.23 ± 0.02	2560 ± 90	HFOSC
2.58 ± 0.05	2820 ± 170	SPRAT
3.59 ± 0.02	2790 ± 350	SPRAT
4.53 ± 0.02	2850 ± 540	SPRAT
5.83 ± 0.05	2210 ± 250	DIS

Download table as:  ASCIITypeset image

X-ray emission from M31N 2008-12a was first detected at a level of (0.6 ± 0.1) $\times \,{10}^{-2}$ ct s⁻¹ on 2016 December 18.101 UT, 5.8 days after the eruption (see Table 4 and also Henze et al. 2016b). Nothing was detected in the previous observation on 2016 December 16.38 UT (day 4.1) with an upper limit of <0.7 $\times \,{10}^{-2}$ ct s⁻¹. Although these numbers are comparable, there is a clear increase of counts at the nova position from the predetection (zero counts in 1.1 ks) to the detection (more than 30 counts in 3.9 ks) observation. Therefore, we conclude that the SSS phase had started by day 5.8.

For a conservative estimate of the SSS turn-on time (and its accuracy), we use ${t}_{{\rm{on}}}=4.9\pm 1.1$ days as the midpoint between days 4.1 and 5.8, which includes the uncertainty of the eruption date. This is consistent with the 2013–2015 X-ray light curves (see Figure 6) for which we estimated turn-on times of 6 ± 1 days (2013), 5.9 ± 0.5 days (2014), and 5.6 ± 0.7 days (2015) using the same method (see HND14, HND15, DHB16). There is no evidence that the emergence of the SSS emission occurred at a different time than in the previous three eruptions.

The duration of the SSS phase, however, was significantly shorter than previously observed (see Figure 6 and Henze et al. 2016c). The last significant detection of X-ray emission in the XRT monitoring was on day 13.9 (Table 4). However, the subsequent 2.9 ks observation on day 15.4 still shows about 4 counts at the nova position, which amount to a 2σ detection (Table 4 gives the 3σ upper limit). Nothing is visible on day 15.9. Again, being conservative, we estimate the SSS turn-off time as ${t}_{{\rm{off}}}=14.9\pm 1.2$ days (including the uncertainty of the eruption date), which is the midpoint between observations 197 and 201 (see Table 4).

In comparison, the SSS turn-off in previous eruptions happened on days 19 ± 1 (2013), 18.4 ± 0.5 (2014), and 18.6 ± 0.7 (2015)—all significantly longer than in 2016. The upper limits in Figure 6 and Table 4 demonstrate that we would have detected each of the 2013, 2014, or 2015 light curves during the 2016 monitoring observations, which had similar exposure times (cf., HND14, HND15, and DHB16). Therefore, the short duration of the 2016 SSS phase is real and not caused by an observational bias.

The full X-ray light curve, shown in Figure 6(a), is consistent with a shorter SSS phase that had already started to decline before day 12, instead of around day 16 as in the last three years. In a consistent way, the blackbody parametrization in Figure 6(b) shows a significantly cooler effective temperature (${kT}=86\,\pm 6$ eV) than in 2013–2015 (${kT}\sim 115\pm 10$ eV) during days 10–14 (cf., DHB16). As previously, for this plot we fitted the XRT spectra in groups with similar effective temperatures.

In contrast to our previous studies of M31N 2008-12a, here our blackbody parameterizations assume a fixed absorption of ${N}_{{\rm{H}}}$  = 0.7 $\times \,{10}^{21}$ cm⁻² throughout. (The X-ray analysis in DHB16 had explored multiple ${N}_{{\rm{H}}}$  values). This value corresponds to the Galactic foreground. The extinction is based on HST extinction measurements during the 2015 eruption, which are consistent in indicating no significant additional absorption toward the binary system, e.g., from the M31 disk (DHG17S; also see DHB16). These HST spectra were taken about three days before the 2015 SSS phase onset, making it unlikely that the extinction varies significantly during the SSS phase. The new N_H, also applied to the 2013–2015 data in Figure 6, affects primarily the absolute blackbody temperature, now reaching almost 140 eV, but not the relative evolution of the four eruptions.

Figure 6(a) also suggests that the SSS phase in 2016 was somewhat less luminous than in previous eruptions. The early SSS phase of this nova has shown significant flux variability; nevertheless, a lower average luminosity is consistent with the XRT light curve binned per Swift  snapshot, as shown in Figure 7. A lower XRT count rate would be consistent with the lower effective temperature suggested in Figure 6(b). Note that this refers to the observed characteristics of the SSS, not to the theoretically possible maximum photospheric temperature if the hydrogen burning had not extinguished early.

Zoom In Zoom Out Reset image size

We show the XRT light curve binned per Swift  snapshot in Figure 7. As found in previous eruptions (HND14, HND15, DHB16), the early SSS flux is clearly variable. However, here the variability level had already dropped by day ~11 instead of after day 13 as in previous years. After day 11, the scatter (rms) decreased by a factor of two, which is significant on the 95% confidence level (F-test, p = 0.03). This change in behavior can be seen better in the detrended Swift  XRT count rate light curve in Figure 7(b). The faster evolution is consistent with the overall shortening of the SSS duration.

The XMM-Newton  light curves from both pointings show clear variability over timescales of a few 1000 s (Figure 8). This is an unexpected finding, since the variability in the Swift  XRT light curve appeared to have ceased after day 11 (in general agreement with the 2013–2015 light curve where this drop in variability occurred slightly later). Instead, we find that the late X-ray light curve around days 14–16 (corresponding to days 18–20 for the "normal" 2013–2015 evolution) are still variable by factors of ~5. The variability is consistent in the EPIC pn and MOS light curves (plotted without scaling in Figure 8).

Zoom In Zoom Out Reset image size

Even with the lower XRT count rates during the late SSS phase, we would still be able to detect large variations similar to the high-amplitude spike and the sudden drop seen in the first and second EPIC light curves, respectively.

The LT eruption spectra of 2016 are broadly similar to those of the 2015 (and prior) eruption (see DHB16), with the hydrogen Balmer series being the strongest emission lines (Figure 9). He i lines are detected at 4471, 5876, 6678 and 7065 Å, along with He ii (4686 Å) blended with N iii (4638 Å). The broad N ii (3) multiplet around 5680 Å is also weakly detected. These emission lines are all typically associated with the He/N spectroscopic class of novae (Williams 1992). The five LT spectra are shown in Figure 9 (bottom) and cover a similar time frame as those obtained during the 2015 eruption. These spectra are also displayed along with all of the other 2016 spectra at the end of this work in Figure 16.

Zoom In Zoom Out Reset image size

The first 2016 spectrum, taken with NOT/ALFOSC 0.54 days after eruption, shows P Cygni absorption profiles on the Hα and Hβ lines. We measure the velocity of the minima of these absorption lines to be at −6320 ± 160 and −6140 ± 200 km s⁻¹ for Hα and Hβ, respectively. This spectrum can be seen in Figure 9 (top), which also shows evidence of a possible weak P Cygni absorption accompanying the He i (5876 Å) line. The first LT spectrum, taken 0.61 days after eruption, also shows evidence of a P Cygni absorption profile on Hα (and possibly Hβ) at $\sim -6000$ km s⁻¹.

This is the first time absorption lines have been detected in the optical spectra of M31N 2008-12a. We note that the HST FUV spectra of the 2015 eruption revealed strong, and possibly saturated, P Cygni absorptions still present on the resonance lines of N v, Si iv, and C iv at t = 3.3 days with terminal velocities in the range 6500–9400 km s⁻¹, the NUV spectra taken ~1.5 days later showed only emission lines (DHG17S).

The HET spectrum taken 1.87 days after eruption can be seen in Figure 10, showing that the central emission profiles of the Balmer lines and He i are broadly consistent. Note that the emission around +5000 km s⁻¹ from the Hα rest velocity probably contains a significant contribution from He i (6678 Å). By this time, the P Cygni profiles appear to have dissipated.

Zoom In Zoom Out Reset image size

Figure 9 clearly shows the existence of high-velocity material around the central Hα line at day 2.58 of the 2016 eruption. This can be seen in more detail, compared to the 2015 eruption, in Figure 10. Note that, as stated above, the redshifted part of the (2016) profile could be affected by He i (6678 Å), although the weakness of the (isolated) He i line at 7065 Å (see Figure 9) suggests that this cannot explain all of the excess flux on this side of the profile. Also note that the extremes of the profile indicate a similar velocity (HWZI ~6500–7000 km s⁻¹).

The 4.91 day spectrum of the 2015 eruption shows Hα and Hβ emission. By comparison, the 2016 4.52 day spectrum also shows a clear emission line from He ii (4686 Å), consistent with the Bowen blend being dominated by He ii at this stage of the eruption. However, we note that this is unlikely to mark a significant difference between 2015 and 2016, as these late spectra typically have very low signal-to-noise ratios (S/Ns). The ARC spectra are shown in Figure 11. The last of these spectra, taken 5.83 days after eruption, shows strong He ii (4686 Å) emission. The S/N of the spectrum is relatively low, but the He ii emission appears narrower than the Hα line at the same epoch, as seen in Figure 12. At this stage of the eruption, we calculate the FWHM of He ii (4686 Å) to be 930 ± 150 km s⁻¹, compared to 2210 ± 250 km s⁻¹ for Hα. The ARC spectra have a resolution of $R\sim 1000$, so these two FWHM measurements are not greatly affected by instrumental broadening. Narrow He ii emission has been observed in a number of other novae. It is seen in the Galactic RN U Sco from the time the SSS becomes visible (Mason et al. 2012). Those authors used the changes in the narrow lines with respect to the orbital motion (U Sco is an eclipsing system; Schaefer 1990) to argue that such emission arises from a re-forming accretion disk. In the case of the 2016 eruption of M31N 2008-12a, we clearly observe the SSS at 5.8 days, meaning this final ARC spectrum is taken during the SSS phase. This is consistent with the suggestion that, in M31N 2008-12a, the accretion disk survives the eruption largely intact (DHG17P). In this scenario, the optically thick ejecta prevent us from seeing evidence of the disk in our early spectra. We note, however, that Munari et al. (2014) argued that in the case of KT Eri, there could be two sources of such narrow He ii emission, initially being due to slower moving material in the ejecta, before becoming quickly dominated by emission from the binary itself (as in U Sco) as the SSS enters the plateau phase.

Zoom In Zoom Out Reset image size

Zoom In Zoom Out Reset image size

DHG17P presented a low-S/N, post-SSS spectrum taken 18.8 days after the 2014 eruption of M31N 2008-12a. This spectrum was consistent with that expected from an accretion disk, and Hβ was seen in emission. However, no evidence of the He ii (4686 Å) line was seen in that spectrum. It is possible that the strong He ii line seen in the ARC spectrum arose from the disk but that the transition was excited by the ongoing SSS at that time.

As with previous eruptions, the emission line profiles of individual lines showed significant evolution during the 2016 eruption. The FWHM of the main Hα emission line (excluding the very high-velocity material) narrows from 4540 ± 300 km s⁻¹ on day 0.54 to 2210 ± 250 km s⁻¹ on day 5.83. The velocity evolution of the 2016 eruption is compared to that of previous eruptions in Figure 10, and is largely consistent. The Hα FWHM measurements of all 2016 eruption spectra are given in Table 8.

The XMM-Newton  EPIC spectra for the two observations listed in Table 6 were fitted with an absorbed blackbody model. The three detectors were modeled simultaneously, with only the normalizations free to vary independently. In Table 9, we summarize the best-fit parameters and also include a simultaneous fit of all EPIC spectra. The binned spectra, with a minimum of 10 counts per bin, are plotted in Figure 13 together with the model curves. The binning is solely used for visualization here; the spectra were fitted with one-count bins and Poisson (fitting) statistics (Cash 1979). The ${\chi }^{2}$ numbers were used as test statistics.

Zoom In Zoom Out Reset image size

Table 9.  Spectral Fits for XMM-Newton  Data

ObsID	${\rm{\Delta }}t$ a	${N}_{{\rm{H}}}$	kT	Red. ${\chi }^{2}$	dof	${{kT}}_{0.7}$ b	Red. ${\chi }^{2}$ b
	(day)	(${10}^{21}$ cm−2)	(eV)			(eV)
0784000101	14.21	${2.2}_{-0.7}^{+0.6}$	${58}_{-5}^{+8}$	1.29	149	${77}_{-3}^{+4}$	1.44
0784000201	15.80	${2.7}_{-0.5}^{+0.6}$	45 ± 5	1.06	140	${68}_{-3}^{+4}$	1.35
Both combined	15.01	2.2 ± 0.4	${53}_{-3}^{+5}$	1.22	291	${73}_{-2}^{+3}$	1.42

Notes.

^aTime in days after the nova eruption (cf., Table 6). ^bThe blackbody temperature (and the reduced ${\chi }^{2}$ of the fit) when assuming a fixed ${N}_{{\rm{H}}}$  = 0.7 $\times \,{10}^{21}$ cm⁻² for comparison with the Swift  XRT temperature evolution (see Figure 6).

Download table as:  ASCIITypeset image

In Table 9 and Figure 13, we immediately see that the two spectra are (a) very similar and (b) contain relatively few spectral counts, leading to a low spectral resolution. The latter point is mainly due to the unexpectedly low flux at the time of the observations, but is also exacerbated by the strong background flaring (cf., Table 6).

In Table 9, we also list a second set of blackbody temperature values (${{kT}}_{0.7}$) for the assumption of a fixed ${N}_{{\rm{H}}}$  = 0.7 $\,\times {10}^{21}$ cm⁻². The purpose of this is to compare these temperatures to the Swift  XRT models which share the same assumption (cf., Section 3.5). In both sets of temperatures in Table 9, there is a slight trend toward higher temperatures in the first observation (day 14.21) compared to the second one (day 15.80). While the binned spectra in Figure 13 give a similar impression, which would be consistent with a gradually cooling WD, it needs to be emphasized that this gradient has no high significance because the two (sets of) temperatures are consistent within their $2\sigma \mbox{--}3\sigma$ uncertainties. In fact, the combined fit in Table 9 has reduced ${\chi }^{2}$ statistics and parameter uncertainties that are similar (the latter even slightly lower) than those of the individual fits.

In Figure 6, the XMM-Newton  data points are added to the Swift  light curve and temperature evolution. For the conversion from pn to XRT count rate, we used the HEASarc WebPIMMS tool (based on PIMMS v4.8d, Mukai 1993) under the assumption of the best-fit blackbody parameters in the third and fourth columns of Table 9.

While the equivalent count rates as well as the temperatures are consistent with the XRT trend of a fading and cooling source, there appear to be systematic differences between the XRT and pn rates. This could simply be due to systematic calibration uncertainties between the EPIC pn and the XRT (Madsen et al. 2017). Another reason might be the ongoing flux variability (see Section 3.6). However, it is also possible that deficiencies in the spectral model are preventing a closer agreement between both instruments. We refrain from an attempt to align the pn and XRT count rates because currently there are too many free parameters (e.g., the potential absorption or emission features discussed in DHB16) and insufficient constraints on them. We hope that a future XMM-Newton  observation will be able to catch this enigmatic source in a brighter state to shine more (collected) light on its true spectral properties.

The precision of the eruption dates for previous outbursts was improved by aligning their light curves, specifically the early, quasi-linear decline (DHB16). For the 2016 eruption, we cannot be certain a priori that this decline phase would be expected to align with previous years because the bright optical peak (Figure 4 left) constitutes an obvious deviation from the established pattern. However, in Figure 2, we find that after the peak feature, most filters appear to decline in the same way as during the previous years. Therefore, we conclude that our estimated eruption date of $\mathrm{MJD}\,=\,57734.32\pm 0.17$ (2016 December 12.32 UT) is precise to within the uncertainties—and this brings about a natural alignment of the light curves.

From the combined optical and X-ray light curves in Figures 2 and 6, it can be seen that in 2016 (i) the optical peak may have been brighter and (ii) the SSS phase was intrinsically shorter than the previous three eruptions (but began at the same time after eruption). In addition, the gap between the 2015 and 2016 eruptions was longer than usual. Below we study these discrepancies in detail and describe them with updated theoretical model calculations. The following discussion ignores the impact of a possible half-year recurrence (cf. HDK15), the potential dates of which are currently not well constrained (except for the first half of 2016; M. Henze et al. 2018, in preparation).

The critical advantage of studying a statistically significant number of eruptions from the same nova system is that we can reasonably assume that parameters like (accretion and eruption) geometry, metallicity of the accreted material, as well as WD mass, spin, and composition to remain (sufficiently) constant. Therefore, M31N 2008-12a  plays a unique role in understanding the variations in nova eruption parameters.

5.2.1. A Brighter Peak After a Longer Gap?

This section aims to understand the surprising increase in the optical peak luminosity (the "cusp") by relating it to the delayed eruption date through the theoretical models of Hachisu & Kato (2006) and Kato et al. (2014, 2017). While the specifics of our arguments are derived from this particular set of models, we note that all current nova light curve simulations agree on the general line of reasoning (e.g., Yaron et al. 2005; Wolf et al. 2013). We also note that DHG17P found an elevated mass accretion rate compared to that employed by Kato et al. (2014, 2017), but again the general trends discussed below do not depend on the absolute value of the assumed mass accretion rate.

The gap between the 2015 and 2016 eruptions was 472 days. This is 162 days longer than the 310 days between the 2013 and 2014 eruptions (see Table 1 and Figure 14) and about 35% longer than the median gap (347 days) between the successive eruptions from 2008 to 2015. The well-observed 2015 eruption was very similar to the eruptions in 2013 and 2014 (DHB16) and did not show any indications that would have hinted at a delay in the 2016 eruption (also see DHG17P). This section compares the peculiar 2016 eruption specifically to the 2014 outburst, because we know that the latter was preceded and followed by a "regular" eruption (see Figures 2 and 6, and DHB16). In general, we know that the peak brightness of a nova is higher for a more massive envelope if free–free emission dominates the SED (Hachisu & Kato 2006).

Zoom In Zoom Out Reset image size

We consider two specific cases. (1) The mean mass accretion rate onto the WD (${\dot{M}}_{\mathrm{acc}}$) was constant but hydrogen ignition occurs in a certain range around the theoretically expected time and, as a result, the elapsed inter-eruption time was longer in 2016 due to stochastic variance. Alternatively, (2) the mean mass accretion rate leading up to the 2016 eruption was lower than typical and, as a result, the elapsed time was longer.

(1) If the mean accretion rates prior to the 2014 and 2016 eruptions were the same, then the mass accreted by the WD in 2016 was ${\rm{\Delta }}{t}_{\mathrm{rec}}\times {\dot{M}}_{\mathrm{acc}}=162\,\mathrm{days}\times 1.6\times {10}^{-7}\,{M}_{\odot }$ yr${}^{-1}\,=0.71\times {10}^{-7}\,{M}_{\odot }$ larger than in 2014. Here we used the mass accretion rate of the $1.38\,{M}_{\odot }$ model proposed for M31N 2008-12a  by Kato et al. (2017). The authors obtained the relation between a wind mass-loss rate and the photospheric temperature (see their Figure 12). The wind mass-loss rate is larger for a lower-temperature envelope, which corresponds to a more extended and more massive envelope.

In Figure 12 of Kato et al. (2017), the rightmost point on the red line corresponds to the peak luminosity of the 2014 eruption. If at this point the envelope mass is higher by $0.71\times {10}^{-7}\,{M}_{\odot }$, then the wind mass-loss rate should increase by ${\rm{\Delta }}\mathrm{log}{\dot{M}}_{\mathrm{wind}}\sim 0.08$.

For the free–free emission of novae, the optical/IR luminosity is proportional to the square of the wind mass-loss rate (see, e.g., Hachisu & Kato 2006). Thus, the peak magnitude of the optical/IR free–free emission is $2.5\times ({\rm{\Delta }}\mathrm{log}{\dot{M}}_{\mathrm{wind}})\times 2\,=\,2.5\,\times 0.08\times 2=0.4$ mag brighter, which is roughly consistent with the increase in the peak magnitudes observed in 2016 in the V and u' bands (Figure 2).

However, the time from optical maximum to ${t}_{\mathrm{on}}$ of the SSS phase should become longer by

$$\begin{eqnarray*}{\rm{\Delta }}t & = & \displaystyle \frac{{\rm{\Delta }}{M}_{\mathrm{env}}}{{\rm{\Delta }}{\dot{M}}_{\mathrm{wind}}+{\dot{M}}_{\mathrm{wind}}}\\ & = & \displaystyle \frac{{M}_{\mathrm{env}}}{{\dot{M}}_{\mathrm{wind}}}\displaystyle \frac{{\rm{\Delta }}{M}_{\mathrm{env}}/{M}_{\mathrm{env}}}{{\rm{\Delta }}{\dot{M}}_{\mathrm{wind}}/{\dot{M}}_{\mathrm{wind}}+1}\sim \displaystyle \frac{6\,\times \,0.35}{0.2+1}=1.75\ \mathrm{days},\end{eqnarray*}$$

where ${M}_{\mathrm{env}}$ is the hydrogen-rich envelope mass. This is not consistent with the ${t}_{\mathrm{on}}\sim 6$ days in the 2016 (and 2013–2015) eruptions.

In general, all models agree that a higher-mass envelope would lead to a stronger, brighter eruption with a larger ejected mass (e.g., Starrfield et al. 1998; Yaron et al. 2005; Hachisu & Kato 2006; Wolf et al. 2013).

(2) For the other case of a lower mean accretion rate, we have estimated the ignition mass of the hydrogen-rich envelope, based on the calculations of Kato et al. (2016, 2017), to be larger by 9% for the 1.35 times longer recurrence period ($0.91\times 1.35=1.23$ yr). Then, the peak magnitude of the free–free emission is $2.5\times ({\rm{\Delta }}\mathrm{log}{\dot{M}}_{\mathrm{wind}})\times 2=2.5\,\times 0.02\times 2=0.1$ mag brighter, but the time from the optical maximum to ${t}_{\mathrm{on}}$ of the SSS phase is longer by only

$$\begin{eqnarray*}{\rm{\Delta }}t & = & \displaystyle \frac{{\rm{\Delta }}{M}_{\mathrm{env}}}{{\rm{\Delta }}{\dot{M}}_{\mathrm{wind}}+{\dot{M}}_{\mathrm{wind}}}\\ & = & \displaystyle \frac{{M}_{\mathrm{env}}}{{\dot{M}}_{\mathrm{wind}}}\displaystyle \frac{{\rm{\Delta }}{M}_{\mathrm{env}}/{M}_{\mathrm{env}}}{{\rm{\Delta }}{\dot{M}}_{\mathrm{wind}}/{\dot{M}}_{\mathrm{wind}}+1}\sim \displaystyle \frac{6\,\times \,0.09}{0.05+1}=0.5\ \mathrm{days}.\end{eqnarray*}$$

The peak brightness of the 2016 outburst is about 0.5 days sooner than that in the 2013, 2014, and 2015 eruptions (see Figure 4 left). These two features, the ~0.1 mag brighter and 0.5 day earlier peak, are roughly consistent with the 2016 eruption except for the ~1 mag brighter cusp (Figure 4 left).

Observationally, we have shown that the expansion velocities of the 2016 eruption were comparable to previous outbursts (Section 4.1). Together with the comparable SSS turn-on timescale (Section 3.5), this strongly suggests that a similar amount of material was ejected. Therefore, scenario (2) would be preferred here.

It should be emphasized that neither scenario addresses the short-lived, cuspy nature of the peak in contrast to the relatively similar light curves before or after it occurred. The models of Kato et al. (2017) and their earlier studies would predict a smooth light curve with brighter peak and different rise and decline rates.

Ultimately, scenario (2) would also require an explanation of what caused the accretion rate to decrease. The late decline photometry of the 2015 eruption indicated that the accretion disk survived that eruption (DHG17P); however, we have no data from 2013 or 2014 with which to compare the end of that eruption. The similarities of the 2013–2015 eruptions would imply that there was nothing untoward about the 2015 eruption that affected the disk in a different manner from the previous eruptions. Therefore, the "blame" probably lies with the donor.

The mass transfer rate in cataclysmic variable stars is known to be variable on timescales from minutes to years (e.g., Warner 1995 and references therein). The shortest period variations (the so-called "flickering"), with typical amplitudes of tenths of a magnitude, are believed to be caused by propagating fluctuations in the local mass accretion rate within the accretion disk (Scaringi 2014). The longer timescale variations that may be relevant to M31N 2008-12a  can cause much larger variations in luminosity. In some cases, as in the VY Sculptoris stars, the mass transfer from the secondary star can cease altogether for an extended period of time (e.g., Robinson et al. 1981; Shafter et al. 1985). The VY Scl phenomena is believed to be caused by disruptions in the mass transfer rate caused by star spots on the secondary star drifting underneath the L1 point (e.g., Livio & Pringle 1994; King & Cannizzo 1998; Honeycutt & Kafka 2004). It might be possible that a similar mechanism may be acting in M31N 2008-12a, resulting in mass transfer rate variations sufficient to cause the observed small-scale variability in the recurrence time and potentially even larger "outliers" as in 2016.

5.2.2. A Shorter SSS Phase

In this section, we aim to explain the significantly shorter duration of the 2016 SSS phase in comparison with previous eruptions and with the help of the theoretical X-ray light curve models of Kato et al. (2017).

While a high initial accreted mass at the time of ignition leads to a brighter optical peak (as discussed in the previous section), it does not change the duration of the SSS phase, assuming that the WD envelope settles down to thermal equilibrium when any wind phase stops. For the same WD mass, a larger accreted mass results in a higher wind mass-loss rate but does not affect the evolution after the maximum photospheric radius has been reached (e.g., Hachisu & Kato 2006). The shorter SSS duration, and thus the shorter duration of the total outburst compared to previous years (Figure 6), therefore needs additional explanation.

Kato et al. (2017) presented a $1.38\,{M}_{\odot }$ WD model with a mean mass accretion rate of $1.6\times {10}^{-7}\,{M}_{\odot }$ yr⁻¹ for M31N 2008-12a. They assumed that the mass accretion resumes immediately after the wind stops, i.e., at the beginning of the SSS phase. The accretion supplies fresh H-rich matter to the WD and substantially lengthens the SSS lifetime, "re-feeding" the SSS, because the mass accretion rate is the same order as the proposed steady hydrogen shell-burning rate of $\sim 5\times {10}^{-7}\,{M}_{\odot }$ yr⁻¹. If the accretion does not resume during the SSS phase, or only with a reduced rate, then the SSS duration becomes shorter. This effect is model independent.

To give a specific example, we calculate the SSS light curves and photospheric temperature evolution for various, post-eruption, mass accretion rates, and plot them in Figure 15. Those are not fits to the data but models that serve the purpose of illustrating the observable effect of a gradually diminished post-eruption re-feeding. The thick solid black lines denote the case of no post-eruption accretion (during the SSS phase). The thin solid black lines represent the case where the mass accretion resumes post-eruption with $1.6\times {10}^{-7}\,{M}_{\odot }$ yr⁻¹, just after the optically thick winds stop. The dashed orange, solid red, and dotted red lines correspond to the mass accretion rates of 0.3, 0.65, and 1.5 times the original mass accretion rate of $1.6\times {10}^{-7}\,{M}_{\odot }$ yr⁻¹, respectively.

Zoom In Zoom Out Reset image size

It is clearly shown that a higher post-eruption mass accretion rate produces a longer SSS phase. Figure 15(a) shows the X-ray count rates in the 2014 (blue crosses) and 2016 (open black circles) eruptions. The ordinate of the X-ray count rate is vertically shifted to match that of the theoretical X-ray light curves (cf., Figure 6). The model X-ray flux drops earlier for lower mass accretion rate, which could (as a trend) explain the shorter duration of the 2016 SSS phase.

Figure 15(b) shows the evolution of the blackbody temperature obtained from the Swift  spectra with the neutral hydrogen column density of ${N}_{{\rm{H}}}$  = 0.7 $\times \,{10}^{21}$ cm⁻² (cf., Figure 6 and Section 3.5). The lines show the photospheric temperature of our models. The model temperature decreases earlier for a lower mass accretion rate. This trend is also consistent with the difference between the 2014 and 2016 eruptions.

Thus, the more rapid evolution of the SSS phase in the 2016 eruption can be partly understood if mass accretion does not resume soon after the wind stops (zero accretion; thick black line in Figure 15). Note that the observed change in SSS duration clearly has a larger magnitude than the models (Figure 15). This could indicate deficiencies in the current models and/or that additional effects contributed to the shortening of the 2016 SSS phase. One factor that has an impact on the SSS duration is the chemical composition of the envelope (e.g., Sala & Hernanz 2005). However, it would be difficult to explain why the abundances of the accreted material would suddenly change from one eruption to the next. In any case, our observations make a strong case for a discontinued re-feeding of the SSS simply by comparing the observed parameters of the 2016 eruption to previous outbursts. The models are consistent with the general trend but need to be improved to be able to simulate the magnitude of the effect.

DHG17P presented evidence that the accretion disk survives eruptions of M31N 2008-12a, the 2015 eruption specifically. In Section 5.2.1, we found that the accretion rate prior to the 2016 eruption might have been lower. If this lower accretion rate was caused by a lower mass transfer rate from the companion, which is a reasonable possibility, then this would lead to a less massive disk (which is potentially less luminous; see M. Henze et al. 2018, in preparation). Thus, even if the eruption itself was not stronger than in previous years, as evidenced by the consistent ejection velocities (Section 4.1) and SSS turn-on timescale (Section 3.5), it could still lead to a greater disruption of such a less massive disk. A part of the inner disk mass may be lost, which could prevent or hinder the re-establishment of mass accretion while the SSS is still active.

This scenario can consistently explain the trends toward a brighter optical peak and a shorter SSS phase for the delayed 2016 eruption. Understanding the quantitative magnitude of these changes, and fitting the theoretical light curves more accurately to the observed fluxes, requires additional models that can be tested in future eruptions of M31N 2008-12a. In addition, we strongly encourage the community to contribute alternative interpretations and models that could help us to understand the peculiar 2016 outburst properties.

5.2.3. Similar Features in Archival Data?

Irrespective of any causal connection between the late 2016 eruption and the newly observed bright cusp, the smooth light curve models cannot explain the nature of this new feature. As the cusp "breaks" the previously smooth presentation of the observed light curve and the inherently smooth nature of the model light curves, it must be due to an additional, unconsidered, parameter of the system. Here we briefly discuss a number of possible causes in no particular order.

The cusp could in principle be explained as the shock breakout associated with the initial thermonuclear runaway, but with evidence of a slower light curve evolution preceding the cusp (see Figure 4, left), the timescales would appear to be incompatible.

An additional consideration would be the interaction between the ejecta and the donor. Under the assumption of a Roche lobe-filling donor, DHG17P proposed a range of WD–donor orbital separations of $25\mbox{--}44\,{R}_{\odot }$; those authors also indicated that much larger separations were viable if accretion occurred from the wind of the donor. Assuming Roche lobe overflow and typical ejecta velocities at the epoch of the cusp of ~4000 km s⁻¹ (see the bottom-right plot of Figure 10), one would expect an ejecta–donor interaction to occur 0.02–0.06 days post-eruption (here we have also accounted for the radius of the donor, $R\simeq 14\,{R}_{\odot };$ DHG17P). With the cusp seemingly occurring 0.65 days post-eruption, the orbital separation would need to be $\sim 330\,{R}_{\odot }$ (~1.6 au). From this we would infer an orbital period in the range of 350–490 days (i.e., $\gtrsim {P}_{\mathrm{rec}}$), depending on the donor mass, and mass transfer would occur by necessity through wind accretion. We note that the eruption time uncertainty (±0.17 days) has little effect on the previous discussion. DHB16, DHG17S, and DHG17P all argued that the system inclination must be low; despite this, it is still possible that the observation of such an ejecta–donor interaction may depend upon the orbital phase (with respect to the observer) at the time of eruption.

As a final discussion point, we note that DHB16 and DHG17S both presented evidence of highly asymmetric ejecta, proposing an equatorial component almost in the plane of the sky and a freely expanding higher-velocity—possibly collimated—polar outflow directed close to the line of sight. We also note that the velocity difference between these components may be a factor of three or higher. If we treat these components as effectively independent ejecta, we would therefore expect their associated light curves to evolve at different rates, with the polar component showing the more rapid evolution. Therefore, we must ask whether the "normal" (2013–2015) light curve is that of the "bulk" equatorial ejecta, and whether the "cusp" is the first photometric evidence of the faster evolving polar ejecta. We note that such proposals have also been put forward to explain multipeak light curves from other phenomena, for example, kilonovae (see Villar et al. 2017 and the references therein).

A consequence of the delayed 2016 eruption is that the dates of the next few eruptions are much more difficult to predict than previously thought. Figure 14 demonstrates how much this surprising delay disrupted the apparently stable trend toward eruptions occurring successively earlier in the year (Section 5.2 discusses the possible reasons).

Currently, detailed examinations of the statistical properties of the recurrence period distribution are hampered by the relatively small number of nine eruptions, and thereby eight different gaps, since 2008 (cf., Table 1). M31N 2008-12a  is the only known nova for which we will overcome this limitation in the near future. For now, we cannot reject the hypothesis that the gaps follow a Gaussian distribution, with Lilliefors (Kolmogorov–Smirnov) test p-value ~0.11, even with the long delay between 2015 and 2016. The distribution mean (median) is 363 days (347 days), with a standard deviation of 52 days. Thereby, the 472 days prior to the 2016 eruption could indicate a genuine outlier, a skewed distribution, or simply an extreme variation from the mean. It is too early to tell.

In addition, all these gaps of roughly one-year length would be affected by the presence of an underlying six-month period, which could dampen the more extreme swings. Of course, the original prediction of a half-year period by HDK15 was partly based on the apparently stable trend toward earlier eruptions since 2008. Comparing this recent trend to the dates of historical X-ray detections in 1992, 1993, and 2001 (HND14), HDK15 found that the most parsimonious explanation for the observed discrepancies between the two regimes would be a six-month shift. However, the putative six-month eruption still remains to be found (M. Henze et al. 2018, in preparation). At present, a single eruption deviating from this pattern does not present sufficient evidence to discard the six-month scenario. The next (few) eruption date(s) will be crucial in evaluating the recurrence period statistics.

While this manuscript was under review, the next eruption was discovered on 2017 December 31 (Boyd et al. 2017). The ~384 days gap between the 2016 and 2017 eruptions is consistent with the pre-2016 eruption pattern. A comprehensive multiwavelength analysis of the new eruption will be presented in a subsequent work.

The 2016 eruption was discovered by Itagaki et al. (2016) using five images (480 s total exposure time) obtained with the 0.5 m f/6 telescope, with a BITRAN BN-52E(KAF-1001E) camera, at the Itagaki Astronomical Observatory, Japan. Additional light curve photometry was first reported in Naito et al. (2016).

The confirmation detection and follow-up photometry of M31N 2008-12a  were obtained at the Half-Meter Telescope of the Xingming Observatory, China, by Itagaki et al. (2016) and Tan et al. (2016). The instrument has a 0.508 m aperture, with a focal length of 2.052 m using a QHY11 CCD camera. All images were calibrated using the standard procedure, including flat-field correction, and dark and bias frames using the Maxim DL software. The relative photometry was obtained in PyRAF with an aperture optimized to the seeing of each individual image. The final magnitudes were calibrated using comparison stars from the XPM catalog (Fedorov et al. 2009).

The central 2k × 2k region of the HFOSC mounted on the 2 m HCT is used for imaging and gives a field of view of ${10}^{{\prime} }\times {10}^{{\prime} }$ at a scale of $0\buildrel{\prime\prime}\over{.} 296$ pixel⁻¹.

Photometric observations were made on 2016 December 14.74 UT in the VRI bands and on December 15.67 in the BVRI bands. The images were bias subtracted and flat-field corrected using twilight flats. Instrumental magnitudes were obtained using aperture photometry. An aperture of radius three times the FWHM was used. Differential photometry was performed with respect to the stars in the field (DHB16) to estimate the magnitude of the nova.

Photometry of M31N 2008-12a  was obtained at the Embry Riddle Aeronautical University, Florida, with (a) a 24 inch CDK Cassegrain telescope equipped with an SBIG STX 16803 detector, and (b) a 1 m RC telescope equipped with an identical detector. Both telescopes took series of 600 s images through the Omega SDSS g', r', and i' filters. The total exposure time for each reported magnitude varied between 1 and 4 hr. The magnitudes were extracted using standard aperture photometric techniques in IRAF (v2.16.1) and calibrated using the DHB16 standard stars in the field. The photometry was first reported in Erdman et al. (2016), Burke et al. (2016), and Kaur et al. (2016).

Late-time optical photometric data was collected with the Danish 1.54 m telescope at the ESO La Silla Observatory, operated remotely from Ondřejov, using the Danish Faint Object Spectrograph and Camera instrument (Hornoch et al. 2016). For each epoch, a series of ten 90 s exposures was taken. Standard reduction procedures for raw CCD images were applied (bias subtraction and flat-field correction) using the APHOT software (Pravec et al. 1994). Reduced images within the same series were co-added to improve the S/N, and the gradient of the galaxy background was flattened using a spatial median filter via the SIPS program. Photometric measurements of the nova were then performed using aperture photometry in APHOT. Five nearby secondary standard stars from Massey et al. (2006) were used to photometrically calibrate the magnitudes included in Table 10.

Table 10.  Complete Data Set of the Visible and Near-infrared Photometric Observations of the 2016 Eruption of M31N 2008-12a

Date	${\rm{\Delta }}t$ a	MJD 57,000+		Telescope and	Exposure	Filter	Photometry
(UT)	(days)	Start	End	Instrument	(s)
2016 Dec 07.110	−5.210	729.085	729.134	AAVSO COO	198 × 15	⋯	>20.2
2016 Dec 09.807	−2.499	731.807	731.834	AAVSO OAR	25 × 60	⋯	>20.3
2016 Dec 10.164	−2.156	732.154	732.174	AAVSO COO	7 × 180	⋯	>18.3
2016 Dec 10.707	−1.599	732.707	732.734	AAVSO OAR	25 × 60	⋯	>20.3
2016 Dec 10.945	−1.375	732.942	732.948	AAVSO BDG	8 × 60	⋯	>19.5
2016 Dec 11.139	−1.181	733.130	733.175	AAVSO COO	6 × 180	⋯	>18.4
2016 Dec 11.338	−0.982	733.330	733.346	Meili 0.4 m	22 × 30	⋯	>19.0a

Note. Includes all observations from $t\sim 7$ days before the eruption until $t\sim 30$ days post-eruption.

^aThe time since eruption assumes an eruption date of 2016 December 12.32 UT.

References. (a) Naito et al. (2016), (b) Itagaki et al. (2016), (c) Tan et al. (2016), (d) Shafter et al. (2016), (e) Burke et al. (2016), (f) Hornoch et al. (2016), (g) Darnley et al. (2016d), (h) Darnley (2016), (i) Erdman et al. (2016), (j) Kaur et al. (2016), (k) Darnley & Hounsell (2016).

Only a portion of this table is shown here to demonstrate its form and content. A machine-readable version of the full table is available.

Download table as:  DataTypeset image

We obtained V-band CCD images with the 1.05 m Schmidt telescope equipped with the Kiso Wide Field Camera (Sako et al. 2012) of the Kiso Observatory, University of Tokyo, Japan. Typical, we took three images with 60 s exposure per night. The dark-subtraction and flat-fielding were performed with IRAF (v2.16.1), before image stacking by using SWarp (v2.19.1 Bertin et al. 2002). Photometry of the stacked images was performed via the aperture photometry package in Source Extractor (v2.8.6; Bertin & Arnouts 1996). We used nearby stars in SDSS, APASS (Henden et al. 2016), and DHB16 for the photometric calibration. The data were first reported in Naito et al. (2016).

Additional g'-, R_C-, and I_C-band upper limits reported by Naito et al. (2016) were obtained using the 0.5 m, f/6.5 MITSuME telescope (Kotani et al. 2005), equipped with an Apogee Alta U6 camera, of the Okayama Astrophysical Observatory, Japan. We took 10 images with 60 s exposure per night for each of the three bands. Image calibration and photometry followed the same procedure as for the Kiso observatory above.

Naito et al. (2016) first reported pre-eruption upper limits and light curve photometry obtained by the 0.51 m, f/12 telescope with the Andor DW936N-BV camera of the Osaka Kyoiku University, Japan. These observations were obtained using an R_C filter with 300 s exposure per image. A stack of 14 images were combined using the IRAF task imcombine. We carried out aperture photometry apphot and PSF photometry daophot within the IRAF environment. Source #11 in DHB16 was used as a comparison star.

Light curve monitoring was performed using a 0.5 m f/6.8 telescope, equipped with an SBIG STL1001E camera, at the Miyaki-Argenteus Observatory, Japan (Naito et al. 2016).

We performed observations at Nayoro Observatory, Nayoro, Japan, using the 0.4 m Meili telescope (Meade Schmidt-Cassegrain Telescope) with an SBIG STL1001E CCD camera (unfiltered or with an R-band filter). The obtained images were reduced in a standard manner and stacked using the StellaImage (v6.5) software. Photometry was conducted using Makali'i, a free software provided by the National Astronomical Observatory of Japan and AstroArts, Inc. for education and research. Magnitudes were measured using an ensemble of comparison stars listed in DHB16. The limiting magnitudes correspond to an S/N of 3 (Naito et al. 2016). The 1.6 m Pirka telescope of the same observatory was also used in this campaign and is described in DHB16.

Additional monitoring data was reported first by Naito et al. (2016) based on remote observations with the following instruments: (a) a 0.5 m f/4.5 CDK astrograph, with an FLI-PL11002M CCD at the New Mexico Skies site (Mayhill, NM), (b) a 0.43 m f/6.8 CDK astrograph plus SBIG STL-11000M CCD at the AstroCamp Observatory hosting site (Nerpio, Spain), and (c) a 0.32 m f/8.0 CDK astrograph, equipped with a SBIG STXL-6303E CCD, at the AstroCamp Observatory.

Pre-eruption upper limits were obtained at Hankasalmi Observatory, Finland using a 0.4 m RC (RCOS) telescope equipped with an SBIG STL1001E CCD. Typically, 25 to 100 unfiltered images with 60 s exposure were obtained per night and stacked using MaxIm DL (v4.61). The stacked image was checked for a nova detection visually by using SAOImage DS9 and photometrically by using a custom software, with an aperture radius of 6'' and a background annulus of 12''–18''. Upper limits were estimated according to the formula $m+2.5\mathrm{log}s/3$, where m and s are the magnitude and S/N, respectively, of comparison star #12 in DHB16.

We observed M31N 2008-12a  with the CBA Concord PF29 telescope—a prime focus 0.74 m f/4.36 reflector on an English Cradle mount—located in suburban Concord, CA. Two cameras were used during this project: an SBIG STL1001E with a clear filter ($1\buildrel{\prime\prime}\over{.} 52$ pixel) and an SBIG STF 8300M (unfiltered, $0\buildrel{\prime\prime}\over{.} 34$ pixel⁻¹, 2 × 2 binning).

Unfiltered groups of 40–50 images of 15 or 20 s duration were taken and median-combined using the AIP4Win⁵⁷ software tool. Typically, two to four sets of these groups were averaged within the AAVSO VPHOT⁵⁸ online photometry tool. The minimum number of subframes was almost always $\gt 100$, usually ~200. The unfiltered or clear-filter measurements were referenced to the V-band comparison stars (see DHB16).

We obtained remote observations with iTelescope.net utilizing (i) the T24 telescope, a Planewave 0.61 m CDK Telescope f/6.5, and an FLI PL-9000 CCD camera at the hosting site in Sierra Remote Observatory, Auberry, CA, and (ii) the T11 telescope, a Planewave 0.5 m CDK with an FLI PL11002M CCD camera at the New Mexico Skies hosting site at Mayfield, NM. Typically, three 5 minute frames (T24) or three 3 minute frames (T11) were obtained in the Luminance filter (a clear filter with UV and IR cutoff). Images were median-combined in AAVSO VPHOT. The detection limits (S/N = 4) are typically 20.7 mag (T24) or 20.2 mag (T11), calibrated using the R-, I-, and V-band standards in DHB16. The photometry was estimated in the same way as for the Concord Observatory above.

We obtained data from the Newcastle Observatory in Newcastle, Ontario, Canada using a 0.40 m Meade Schmidt-Cassegrain (ACF) Telescope working at f/7 and a QSI 516ws CCD camera. The images were obtained in the filters Johnson V and Cousins I_C, or unfiltered. Most images were a stack of six frames, median-combined to minimize cosmic-ray effects. At least one imaging run was obtained per night when weather permitted. Occasionally, a second imaging run in the same night (before dawn) was attempted.

All individual images were automatically put through an image-processing pipeline for bias, dark, and flat-field calibration, as well as plate-solved to include WCS coordinates before being stacked for analysis. The stacked image is viewed in Aladin (v9; Bonnarel et al. 2000; Boch & Fernique 2014) with the SIMBAD database (Wenger et al. 2000) loaded to accurately locate the target. The detection of the target was compared using comparison star #8 in DHB16 which has a V-band magnitude of 19.087. If the target was not detected (S/N < 3), the limit was reported as fainter than 19.1 mag.

We obtained optical images and photometry of M31N 2008-12a  on 2017 January 08.12 UT with the 8.4 m LBT and Multi-Object Double Spectrograph (MODS2). Images were obtained in the standard SDSS u'g'r'i'z' filters with a total integration time of 300 s in each of the g'r'i'z' filters and 600 s in the u' filter at an image scale of 0125 per pixel with a field of view of about $6^{\prime} \times 6^{\prime}$. Image quality was typically 08 to 10 under nonphotometric conditions. Bias and twilight-sky flat-field images were obtained in each of the u'g'r'i'z' filters to facilitate the data reduction. All reductions were performed using IRAF (v2.16).

We obtained observations at West Challow Observatory, Oxfordshire, UK, on most clear nights using a 0.35 m Meade Schmidt-Cassegrain Telescope working at f/6.3 with a clear filter and a Starlight Xpress SXVR-H9 CCD camera. Typically, 20–30 images with 60 s exposure were recorded, dark-subtracted, flat-fielded, and stacked using Astrometrica. Having determined that the nova was not visible in the stacked image at the expected position, the magnitude of the faintest clearly recognizable stellar object in the vicinity of the nova as determined by Astrometrica was reported as the faint magnitude limit for that night. When detected, the magnitude of the nova was measured using the AIP4WIN software⁵⁹ and an ensemble of the V-band comparison stars listed in DHB16.

Light curve photometry was obtained at Bernezzo Observatory, Italy, using a 0.25 m f/4 reflector with an Atik 314L CCD and a scale of $1\buildrel{\prime\prime}\over{.} 33$ per pixel. We stacked 19 individual V-band images with 120 s exposure each for an S/N = 29.6 detection listed in Table 10. The astrometric solution was calibrated through the Astrometrica software. The photometry was extracted using the software FotoDif (v3.95)⁶⁰ and calibrated via the AAVSO Variable Star Plotter,⁶¹ which uses comparison starts from DHB16.

We observed the nova using a 0.4 m SCT telescope, equipped with an SBIG 6303 CCD, located on Haleakala, Hawaii, as part of the LCO group.⁶² The exposure times were 60 s with no stacking, and flats and darks were applied by LCO. The filter was Johnson V, and the photometry used was AIP4WIN using the aperture function. The calibration stars were taken from the APASS catalogue (Henden et al. 2016).

We obtained light curve photometry using a Meade 0.4 m SCT, with an Astrondon V-band photometric filter and an FLI Proline CCD camera (with 1k × 1k back-illuminated SITe chip), located at New Smyrna Beach, FL. Typically, 20–30 sets of 60 s exposures were stacked. The image capture and photometry used the Maxim DL v6.14 software. The photometry was calibrated using the comparison starts from DHB16.

Images were obtained at the Polaris Observatory, Budapest, Hungary, using a 0.25 m f/4 Newtonian reflector with a V filter and an ALCCD5.2 (QHY6) CCD camera. All raw images were processed with gcx v1.3 (dark-subtraction and flat-field correction and stacking). The integration times were 12 × 180 s. The stacked image was plate-solved with the solve-field tool of astrometry.net. The aperture photometry was performed using IRAF (v2.16.1) and calibrated using the V-band reference stars of DHB16 via the AAVSO VSP.⁶³

One set of two 400 s Hα images (central wavelength 6600 Å; FWHM 145 Å) was obtained during the eruption with the JAST/T80 telescope at the Observatorio Astrofisico de Javalambre, in Teruel, owned, managed, and operated by the Centro de Estudios de Fisica del Cosmos de Aragon. The aperture photometry was derived using the Source Extractor software (v2.8.6; Bertin & Arnouts 1996) and calibrated with R-band data of the Local Group Galaxies Survey (Massey et al. 2006).

Light curve photometry was obtained via remote observations at the ROTAT and SATINO-2 telescopes, both located at the Observatoire de Haute Provence, France. The telescopes are remotely operated by the "Foundation Interactive Astronomy and Astrophysics," Germany. ROTAT is a 0.60 m f/3.2 Newtonian reflector used with a clear filter and an SBIG 11000 STL CCD camera. SATINO-2 is a 0.30 m f/6 Schmidt-Cassegrain reflector used with a clear filter and an SBIG ST8-E CCD camera. ROTAT photometry is estimated from a calibrated 600 s guided exposure, while SATINO photometry is based on 19 calibrated and summed 300 s exposures. The photometric analysis was carried out with the MIRA PRO x64 software (v8.012). The photometry was calibrated using the R-band magnitudes of the DHB16 comparison stars #11 and #12.