ejected (Prialnik, Shara & Shaviv 1979 ). Once accretion resumes the WD's envelope grows until it again erupts as a nova. All novae must erupt thousands of times o v er the course of their multi-Gyr lifetimes (Ford 1978 ).

The time between successive nova eruptions depends strongly on the mass of the accreting WD. Suites of nova models (Yaron et al. 2005 ) demonstrated that low mass (0.6-0.8 M ) WDs with relatively large radii and weak surface gravities must accumulate massive envelopes (10 -3 -10 -4 M ) to generate enough pressure to initiate nova TNRs. This can take a Myr or more, so that nova eruptions on low mass WDs are expected to be relatively rare, as is observed (Shara et al. 2018 ). More massive WDs ( ∼ 0.8-1.1 M ) with higher surface gravities can accumulate critical-mass envelopes which yield TNRs in just thousands of years (Yaron et al. 2005 ). Novae containing these more massive WDs are the ones seen to erupt most frequently in an y giv en few centuries (Shara et al. 2018 ), the time-span in which humans have been detecting and recording nova eruptions. WDs with masses in excess of ∼1.2 M can accumulate critical- The time between successive nova eruptions also depends on the average rate of mass transfer d M / d t in a pre-nova binary. Higher d M / d t will more rapidly bring an accreted envelope to the critical mass (for a given WD mass) to initiate a nova TNR. The larger radii and smaller surface gravities of subgiant and giant donors enable them to transfer mass faster than main sequence stars. It is thus not surprising that nearly Chandrasekhar mass WDs with post main sequence companions are the ideal recipe for making recurrent novae. Pagnotta & Schaefer ( 2014 ) pointed out that recurrent novae tend to display modest outburst amplitudes, long orbital periods, and red infrared colours (all due to their post main sequence donors). In addition, the y e xhibit high mass ejection v elocities, high e xcitation spectral lines, an eruption light curve with a plateau, and a WD mass greater than 1.2 M (all due to their massiv e WDs). The y then singled out several likely recurrent novae masquerading as 'ordinary' novae, including KT Eri (see Section 1.5).

The most rapidly recurring nova known (M31N 2008-12a, hereafter simply '12a') is situated in the Andromeda galaxy. Its annual nova eruptions (Darnley et al. 2014 ) must be due to a nearly Chandrasekhar-mass WD (Hillman et al. 2016 ), which needs to accrete just ∼10 -7 M to trigger a TNR. The donor is either a low luminosity red giant or a horizontal branch (red clump) star (Darnley et al. 2017 ) which is capable of transferring mass to its WD at a rate d M / d t that is high enough ( ∼ 10 -7 M yr -1 ) to enable annual nova eruptions. The powerful TNRs on the WD of 12a lead to mass ejection near the beginning of the outburst with velocities as high as 13 000 km s -1 (Darnley et al. 2016 ), which then decrease throughout an outburst (Darnley et al. 2017 ).

An inevitable conclusion of the observed range of ejection velocities in 12a is that the early, fast ejecta of the (N + 1)th eruption must o v ertake, collide and mix with the late, slower ejecta of its Nth eruption. In addition, the outermost ejecta must have encountered and swept up surrounding ISM, and thus been decelerated, making it a slow-moving target for later eruptions' ejecta. There is evidence for this phenomenon in the ejecta of the recurrent nova T Pyx (Schaefer, Pagnotta & Shara 2010 ), where previously faint knots have 'turned on'.

Ground based and Hubble Space Telescope images have revealed a vast ring-like structure of ejecta and swept-up ISM surrounding 12a (Darnley et al. 2019 ). At 134 pc in size it is larger than most supernova remnants, and has been dubbed the first (and to date only known) nova 'super-remnant'. The phenomenon of successive ejecta colliding with each other and with surrounding ISM should describe not just 12a, but all RNe (Healy-Kalesh et al. 2023 ). These authors' e xtensiv e simulations demonstrated that nova super-remnants encompassing RNe should be ∼20-120 parsecs in extent. If this prediction is correct, then why is the 12a super-remnant the only one known?

Depending on the mass ejected and the density of the surrounding ISM, the shells of novae erupting infrequently should fade on timescales of decades to multiple centuries (Tappert et al. 2020 ) after eruption. The shells of GK Per (nova Per 1901 CE ) (Duerbeck & Seitter 1987 ;Shara et al. 2012a ) and Nova Sco 1437 CE (Shara et al. 2017 ) remains easily detectable with 1-m-class telescopes, while the much fainter 2000 + year old ejecta of Z Cam is only partially revealed in multihour exposures with 4-m class telescopes (Shara et al. 2007(Shara et al. , 2012b ) ). Much deeper imaging, designed to reach the lowest surface brightnesses ever achieved for nova ejecta, have 'filled in' the previously 'missing' sections of Z Cam's shell, and hint at a second larger, concentric and still-fainter shell (Shara et al. 2023 ) o v er 1 de gree in diameter. We posit that muc h more sensitive and lar ger ar ea imaging than has yet been accomplished is essential to detecting most hitherto unseen no va ejecta, including no va superremnants, due to their extremely low surface brightnesses. A first test of this hypothesis is the essence of this paper.

KT Eridani (Nova Eridani 2009) was a fast nova (fading from peak brightness m ∼ 5.4 by 3.0 mag in 13.6 d), whose light curve was serendipitously captured by the Solar Mass Ejection Imager (SMEI) instrument on the Coriolis satellite in exquisite detail (Hounsell et al. 2010 ). Pagnotta & Schaefer ( 2014 )'s suggestion that KT Eri might be a recurrent nova was strengthened by the detailed analysis of Schaefer et al. ( 2022 ). They determined an orbital period of 2.61595 d and a companion star temperature of 6200 ± 500 K, hence a subgiant companion; an absolute system V magnitude of + 0.7 ± 0.3, implying a very high d M / d t of 3.5 x 10 -7 M yr -1 ; light and radial v elocity curv es that demonstrate a white dwarf mass of 1.25 ± 0.03 M ; and a likely recurrence time-scale ∼50 yr (though only one eruption has been recorded to date).

That a nova which reaches naked-eye brightness at maximum light could have been missed in the 1950s or 1960s might seem surprising. But as emphasized by Pagnotta & Schaefer ( 2014 )...'Eruptions o v er the last century can easily be missed if they happen when the star is too close to the Sun, during the full moon, during any of many intervals when no one was watching, or if the search did not go deep enough even if someone was searching in the right area at the right time. For CNe, the discovery efficiency is 22 per cent for novae peaking at V = 6 mag and 9 per cent for novae peaking at V = 10 mag, even in ideal conditions; this rate remains fairly constant from 1890 to 2012.' That previous eruptions of KT Eri, early in and during the middle of the 20th century, have likely been missed is not surprising at all. At a Gaia -determined distance (Bailer-Jones et al. 2021 ) of ∼5 kpc, the predicted ∼20-120 pc diameter shell surrounding KT MNRAS 529, 224-235 (2024) 6. SALT spectrum of the outer KT Eri shell. A few sources are background galaxies. A few short horizontal line segments (especially on H α) are artefacts due to imperfectly remo v ed cosmic rays. All faint unidentified lines are due to the night sky.

Condor consists of six apochromatic refracting telescopes of objective diameter 180 mm, each equipped with a large-format (9576 × 6388 pix els), v ery low read-noise (1.2 e -1 ), very rapid readtime ( < 1 s) CMOS camera. Though the six telescopes are aligned on the same point in the sk y, the y hav e no interferometric capability. In addition to Sloan g ', r ', and i ' filters, this 'array' telescope is equipped with a set of narrow-band filters (each of 3 nm FWHM, one per telescope). The ions (central wavelengths) corresponding to those six filters are He II (468.6 nm), [ O III ] (500.7 nm), He I (587.6 nm), H α (656.3 nm), [ N II ] (658.4 nm), and [ S II ] (671.6 nm).

Condor is located at the Dark Skies New Mexico observatory near Animas, New Mexico. A full description of the telescope is given in Lanzetta et al. ( 2023a , b ).

We used Condor with its complement of narrow-band filters to observe KT Eri between 2021 October and 2022 November. The Condor CMOS cameras obtain an image scale of 0.85 arcsec per pixel, which results in a 134 × 91 arcmin 2 field of view. Each exposure was 600 s in length. The observation log, filter central wavelengths and the 'reach' = the net observation time multiplied by the aperture area through each filter (a useful metric for comparing array telescopes) are summarized in Table 1 . The array was dithered by a random offset of 15 arcmin between exposures. Images of the dusk and dawn twilight sky were obtained every night, and bias observations were obtained at the end of every night.

The observations were processed through the Condor data pipeline (Papers I and II), with steps involving bias subtraction, flat fielding and background subtraction, astrometric calibration, and photometric calibration. As described in Lanzetta et al. ( 2023a ), the astrometric calibration yields systematic uncertainties of ࣠ 0.1 arcsec. Difference images were constructed using the saccadic fast Fourier transform

Two spectra of the KT Eri shell, each one hour in length, were obtained with the Southern African Large Telescope (SALT; O' Donoghue et al. 2006 ) and its Robert Stobie Spectrograph (Burgh et al. 2003 ;Kobulnicky et al. 2003 ) using its PG2300 grating on 2023 February 22 and 23. The grating/spectrograph co v ered the wavelength range 6510-6725 Å with a resolution of 2.2 ± 0.25 Å. The data were reduced with the SALT RSS pipeline described in Kniazev ( 2022 ), which includes correction of bad columns and lines on the CCDs; calculation and application of gain-correction coefficients; construction of a spectral flat field for subsequent pixel sensitivity correction; and location and removal of cosmic ray events. These steps are followed by reduction of a reference spectrum; reduction of a spectrophotometric standard, and reduction of the object.

The Galactic coordinates of KT Eri are (207.9863 • , -32.0202 • ). At a Gaia distance of 5110 + 920 -430 pc from adopting a Bayesian prior on the stellar density along the line of sight (see Schaefer et al. 2022 ), KT Eri lies ∼3 kpc below the Galactic Plane. Its radial velocity relative the Sun is -142 ± 5 km s -1 (Schaefer et al. 2022 ). KT Eri lies in the direction of the Orion-Eridanus superbubble (Heiles 1998 ;Boumis et al. 2001 ;Frisch 2007 ), but far beyond it. The superbubble is a set of o v erlapping superno va remnants which may be associated with the Orion OB1 association, centered ∼800 pc from the Sun. Superbubbles and their associated supernov ae gi ve rise to Galactic chimneys (Norman & Ikeuchi 1989 ) and hot gas in the MNRAS 529, 224-235 (2024) Galactic halo (Henley et al. 2010 ). Hill et al. ( 2012 ) modeled the time-dependent, supernov a-dri ven, turbulent, magnetized interstellar medium below the Galactic Plane. They found that warm ( T ∼10 5 K) and hot gas ( T > 10 6 K) with densities of ∼10 -3 -10 -4 cm -3 alternately dominate at | z | ∼ 3 kpc, and that large-scale vertical oscillations of the chimney gas are ongoing. Differential rotation of the Galactic disc, and gas heating mechanisms in addition to superno vae (Henle y et al. 2015 ) further complicate any effort to predict the velocity of gas, relative the Sun, in the neighborhood of KT Eri.

The six Condor narrow-band images of KT Eri are shown as a mosaic in Fig. 1 , with KT Eri indicated by a pair of red tick marks. The coherent, shell-like structure surrounding KT Eri is strongest by far in H α, where the typical surface brightness is 4 × 10 -17 erg s -1 cm -2 arcsec -2 .

Much fainter patterns of nebulosities are seen in each of the three images of He II , [ O III ], and He I . The large range of ionization potentials of these three ions suggests that we are not detecting their emission lines in our narrow-band images, but rather continuum light scattered by Galactic cirrus. There are much weaker indications of [ N II ] and [ S II ] filter -image neb ulosities that mimic the pattern seen in H α, so we cannot rule out that these are line-emitting regions. But the very weak [S II ] argues against the nebulosity in Fig. 1 being due to a supernova remnant.

The five difference images (H α minus the other five filters), as well as H α minus the (appropriately scaled) sum of the other five narrow-band images, are shown in Fig. 2 . It is again apparent that faint nebulosity pervades the entire region, and that the ∼0.6 deg diameter structure surrounding KT Eri is (by far) most apparent in H α.

In the top panel of Fig. 3 we again show the H α -[ N II ] image. The tw o filters' w avelengths are separated by just 2.1 nm, giving the best subtraction of stars and continuum light possible. The Gaia propermotion vector is also shown, indicating KT Eri's motion towards the SE. The lower panel of Fig. 3 shows three brightness contours (red, cyan, and yellow contours corresponding to H α -[ N II ] flux density differences of 0.25, 0.40, and 0.55 µJy, respectively).

The angular size of the shell, if at the ∼5 kpc distance of KT Eri, corresponds to ∼50 pc. This is comparable (logarithmically) in size to the 134 pc super-remnant of 12a, but two orders of magnitude larger than the ∼0.1-1 pc-sized ejecta of 19th and 20th century novae. KT Eri last erupted as a nova in 2009, and the speed of its ejecta was > 4000 km s -1 (Yamaoka et al. 2009 ). Even ignoring deceleration as those ejecta interact with the surrounding interstellar medium (Duerbeck & Seitter 1987 ), it will take the ejecta ∼6000 yr to achieve the observed size of the KT Eri shell.

As shown in the accompanying paper (Healy-Kalesh et al. 2024 ) which models the interactions of KT Eri's successive ejecta with each other, the currently observed shell can be grown to its observed size in no less than 25-51 k yr. Giv en the Gaia -deriv ed parameters of KT Eri, we are able to contextualize its position and motion relative the Earth and Galactic Center o v er the time-scale of the shell's formation. In particular, we can pose and answer the question: was KT Eri inside its nebulosity 50 kyr ago, as must have been the case if the nova is responsible for the nebulosity now surrounding it? Using the GALA galactic dynamics package (Price-Whelan 2017 ; Price-Whelan et al. 2020 ) and a multi-component Milky Way potential model with disk parameters from Bo vy ( 2015 ), we inte grated the orbit of KT Eri backwards 51 kyr from Gaia 's 2016.0 reference epoch. Over this time, KT Eri moved 14.3 pc with respect to the Galactic Center, or approximately 0.29 shell diameters.

To determine the sensitivity of the computed displacement of KT Eri to the adopted orbital parameters, we again adopted the Bovy model for the Galactic potential. We then computed 800 random orbits for KT Eri which sampled the multidimensional Probability Distribution Function assuming Gaussian errors in proper motions, including the covariance in RA and Dec., radial velocity and distance. For the distance to KT Eri we used a uniform distribution between 5110 + 920 and 5110-430 pc. Looking back 51 kyr gives a median distance traveled by the binary of 18-20 pc, somewhat larger than abo v e, but still well within the bounds of the observed shell. This result demonstrates that (1) the likely uncertainty in how far KT Eri has traveled since its putative began forming is just ∼3-5 pc.

In summary, KT Eri's continued presence within its own shell is consistent with the shell's ejection history and the binary's motion within the Galaxy.

The locations of the two slit positions used in our RSS observations are shown in Fig. 4 , and the reduced 2D spectra are shown in Figs 5 and 6 . The faint, curved features between ∼660.0-662.5 nm in the 2D spectra are residual artefacts due to slightly imperfect correction of the gain of the two amplifiers on each of the RSS CCD detectors. The slight 'wiggle' in the [S II ] line at 671.6 nm in Fig. 5 is also an artefact. The few point sources are background galaxies.

The profiles of the d α line in the inner and outer part of the shell, compared to the nearest night-sky OH line, are shown in Figs 7 and 8 , respectively. The H α line is the only line resolved in the spectra. It is broader than any of the other lines, all of which are due to the night sky, including [ N II ] and all of the OH lines. H α 656.28, OH 653.30, and [ N II ] 658.35 nm have full widths at zero intensity of approximately 0.55, 0.45, and 0.45 nm, respectively. This demonstrates that there is H α emission with velocity components displaced by up to ±125 km s -1 with respect to the Sun in the KT Eri nebular spectra. There is no velocity gradient in the d α line across the nebula. The lack of higher velocities (multiple hundreds or thousands of km s -1 ) argues (as does the lack of strong [S II ] emission) against the nebulosity of Fig. 1 being due to a supernova remnant.

The simplest shell model is spherical, uniform and uniformly expanding. It would be expected to yield a combined spectral profile of one central peak (the bright edges of the shell) with symmetric broad wings (the rest of the shell). The centroid, spacing, and breadth of these expected features are dictated by the velocity of the shell with respect to Earth, its expansion rate and turbulent peculiar velocities within the shell's leading edges.

At the radial velocity of KT Eri ( -142 ± 5 km s -1 ), the rest wavelength of H α emission is blue-shifted by 0.31 nm to 656.97 nm. If the shell mass centroid shares KT Eri's space velocity, the line-ofsight components would appear up to ±0.55 nm symmetrically about 655.97 nm, assuming a (model-dependent) maximum expansion velocity of 250 km s -1 (Healy- Kalesh et al. 2024 ). In this scenario, these shell components would be obscured by the OH 655.36 nm and H α 656.28 nm lines in the inner slit spectrum. If the shell shares the