COMPREHENSIVE PHOTOMETRIC HISTORIES OF ALL KNOWN GALACTIC RECURRENT NOVAE

Recurrent novae (RNe) are ordinary novae systems for which the recurrence time scale happens to be less than roughly one century. That is, RNe are binary stars where matter accretes from a donor star onto the surface of a white dwarf, where the accumulated material will eventually start a thermonuclear explosion that makes the nova eruption. The ordinary nova systems for which only one eruption has been detected are labeled as "classical novae" (CNe), even though some of them might really be RNe for which all but one eruption from the last century has been missed. The nova systems have a continuum of recurrence time scales, ranging from roughly a decade to perhaps 100,000 years or longer, with the division at roughly one century being purely based on observational expedience.

Why do some nova systems have a recurrence time scale(τ_rec) of less than a century? Here, theory provides a confident answer that two conditions are required to get short inter-eruption times. The first requirement is that the white dwarf mass must be near the Chandrasekhar mass. The reason is that the high white dwarf mass makes for a high surface gravity which allows even a relatively small amount of matter to reach the kindling temperature for thermonuclear runaway. The limit is that the mass of the white dwarf (M_wd) must be greater than something like 1.2 M_☉. The second requirement is that the mass accretion rate ($\dot{M}$) must be high. The reason is that a high $\dot{M}$ is the only way to get enough material onto the white dwarf to reach the runaway condition during the short inter-eruption time interval. Indeed, the RNe have an accretion rate higher than most cataclysmic variables (CVs), with values close to 10⁻⁷ M_☉ yr⁻¹ (Hachisu & Kato 2000, 2001; Webbink et al. 1987; Patterson et al. 1998; Knigge et al. 2000; Hachisu et al. 2000b, 2003).

So the RNe have white dwarfs near the Chandrasekhar mass and matter is being piled on at a very high rate. With this situation, it is possible that the white dwarf will soon be driven to the Chandrasekhar mass and then collapse as a Type Ia supernova. As such, RNe are a likely answer to the perennial Type Ia progenitor question. Often in the literature, discussion does not go past this simple argument. But there is a major loophole in that the mass ejected during each nova event has not been accounted for. If the white dwarf ejects more mass during an eruption than it has accreted during the previous inter-eruption interval, then the white dwarf will not be gaining mass and will not collapse as a Type Ia explosion. So the question really comes down to evaluating the recurrence time scale(τ_rec), the mass accretion rate ($\dot{M}$), the mass ejected by the nova event (M_ejecta), and then whether $M_{\rm ejecta}>\tau _{\rm rec}\dot{M}$. It turns out that the ejecta mass has a very large uncertainty (over two orders of magnitude) from both observational and theoretical work. Also, my recent work (see below) has demonstrated that τ_rec generally has errors of a factor of 3 and $\dot{M}$ has large uncertainties. Until we can answer the question of whether $M_{\rm ejecta}>\tau _{\rm rec}\dot{M}$, RNe can only remain a good idea as a solution for the progenitor problem.

Another potential problem for the RNe answer is whether there are enough RNe in our Galaxy and Local Group to produce the Type Ia events. That is, is the RN death rate equal to the Type Ia supernova rate? Published answers to date appear to say "no" both from a theoretical population synthesis (Branch et al. 1995) and an observational basis (della Valle & Livio 1996). (Unfortunately, the answer appears to be "no" for all candidate classes of progenitors; see Branch et al. 1995.) But the theoretical basis for the "no" to RNe is suspect as population synthesis analysis has very large uncertainties when dealing with rare evolutionary paths, especially after common envelope binary evolution has occurred. This distrust of population synthesis for evaluating rates in such cases is further emphasized by studies of the fraction of 3–8 M_☉ stars that must undergo Type Ia explosions is orders of magnitude higher than theory would give (Maoz 2008). And the observational "no" is also suspect because the realities of discovery statistics were ignored and these can easily increase the RN death rate by a factor of 100 or more. The way to answer the RNe death rate question can only be to get better demographics for RNe (including their numbers, light curves, accretion rates, and recurrence time scales) as well as to determine their real discovery statistics.

The Type Ia progenitor question is an old one, with major reviews from each of the last four decades appearing in Livio (2000), Branch et al. (1995), Trimble (1984), and Whelan & Iben (1973). Even today, extensive debates rage as to which of many systems account for most of the Type Ia events. The primary dividing line is whether the progenitors have one or two white dwarfs in the binary. The double degenerate idea has two white dwarfs in close orbit for which angular momentum loss by gravitational radiation grinds the orbit down until the two white dwarfs collide and exceed the Chandrasekhar mass resulting in a supernova explosion. No double white dwarf system with adequate combined mass and a short enough orbital period has ever been identified, despite massive efforts, so we cannot point to any star in the sky as being a progenitor of this class (Robinson & Shafter 1987; Napiwotzki et al. 2001). For example, even the star KPD 1930+275 with a white dwarf in a binary of total mass 1.47 ± 0.01 M_☉ (Maxted et al. 2000, but see Geier et al. 2007) will lose enough mass so as to avoid exceeding the Chandrasekhar mass (Ergma et al. 2001). The single degenerate model has a mass donor star feeding matter onto a white dwarf which eventually exceeds the Chandrasekhar mass also resulting in a supernova explosion. For this model, we can point to several populous classes of stars which are candidate progenitors including the RNe, symbiotic stars, and super-soft sources. These classes have substantial overlap within their populations. With these many possibilities, all of which presumably work, the question comes down to which is the dominant progenitor. And with the possibility of two roughly equal populations of Type Ia supernovae (Sullivan et al. 2006), there might even be two dominant progenitor classes.

The progenitor question has recently come to the forefront as being of high importance. The reason is that Type Ia supernovae have become a premier tool for cosmology, for example, with the Hubble diagram providing the first and strongest evidence for the existence of Dark Energy. But supernova cosmology is based on the idea that the luminosity–decline relation has no evolution with redshift. If z ∼ 1 supernovae systematically have lower metallicity (appropriate for their long look-back time) then their envelopes will have systematically lower opacity with an inevitable change in the luminosity–distance relation and a systematic distortion of the derived distances. Such systematic effects will lead to a change in the shape of the Hubble diagram, with this error perhaps being interpreted as evidence for Dark Energy. Domínguez et al. (2001) have demonstrated that the range of evolutionary effects (from systematic changes in redshift of both the metallicity and the main-sequence mass of the progenitor) can lead to typical errors of ∼0.2 mag out to z = 1. The supernova researchers only argue against evolution with their evidence that the high-z events have similar spectra to low-z events (Hook et al. 2005), although this is not proof of the lack of evolution at less than the 0.2 mag level. The question of evolution is critical as the amount of evolution at z ∼ 1 is comparable to the differences between cosmologies with and without Dark Energy. Without knowing the identity of the progenitor, evolution calculations are not possible and the effect can significantly change the shape of the Hubble diagram. So, in principle, the progenitor problem is critical for the entire supernova cosmology enterprise. And it will only become more important, as future experiments (like the SNAP satellite) are seeking to measure the Hubble diagram with an accuracy of 0.01 mag, a level at which evolution effects can dominate and perhaps make the goal unreachable. So a solution to the progenitor problem has wide importance.

The solution can come only with a broad investigation of the various candidate progenitors. A part of this will be the study of RNe, with particular attention to questions relating to whether $M_{\rm ejecta}>\tau _{\rm rec} \dot{M}$ and the value for the RN death rate. These studies are essentially demographic investigations that must examine many fine details of all the known RNe systems.

The only comprehensive review or summary of RNe is the influential and superb paper by Webbink et al. (1987). But a lot has been learned since that time. Unfortunately, all of the many subsequent observations are widely scattered. For later reviews, we have a conference proceeding on one specific eruption (Evans et al. 2008), two brief overviews in conference proceedings (Anupama 2002, 2008), and one short summary (Sekiguchi 1995). A deeper problem is that all of the photometry from before 1970 or so has systematically wrong magnitudes due to universal errors in the magnitude scale, and thus the majority of RNe events have only poorly measured properties.

Over many years, I have been collecting a large database of observations of all the known galactic RNe, both in quiescence and in eruption. For example, I have measured comparison star magnitudes and sequences so that old and new magnitudes can be correctly placed onto a uniform modern magnitude scale. I have also remeasured all the archival photographic plates of all the pre-1970 eruptions, so that the light curves are directly comparable. I have made exhaustive searches of the archival material for previously undiscovered eruptions. I have extensive long-term light curves for the RNe at quiescence. I have 47 and 80 eclipse times on both sides of the latest eruptions of U Sco and CI Aql, respectively. The easy-to-point-to results are the discovery of one new RN system, the identification of six previously unknown eruptions, the discovery of five of the orbital periods, the confident measure of the orbital period change across two eruptions, and the prediction of the dates of the next eruptions of U Sco and T Pyx.

Now is the time for a comprehensive review of RNe photometry because of (1) the importance of the progenitor problem, (2) the lack of recent comprehensive reviews, (3) the scattered nature of the data in the literature, (4) the now-corrected calibration problems with the old magnitudes, and (5) the new data. This paper seeks to fill the need by providing a comprehensive set of new data, remeasured old data, and data from the literature. The goal is to provide a uniform analysis on the various properties of all known galactic RNe and the demographics of the group.

What are the known RNe in our own Galaxy? This is an important question because the best census is needed for valid demographic studies. The question of which stars are RNe has sometimes been uncertain. One source of confusion is whether a particular eruption is a nova event (involving thermonuclear runaway on the surface of the white dwarf) or a dwarf nova event (involving a brightening of the accretion disk). Dwarf novae with rare high-amplitude eruptions can be easily confused with RNe, as can CNe that also display dwarf nova eruptions.

There is a long history of claims for stars being RNe. One of the first discovered variable stars is Mira, and it was called a nova when its first peak was recognized and then a second eruption was spotted. The modern idea of "RN" had to await the separation of nova events as based on spectroscopy. At various times, several dwarf novae (including WZ Sge, VY Aqr, RZ Leo, and V1195 Oph) have been identified as RNe (Webbink et al. 1987). The two eruptions of V616 Mon (A0620-00) pointed to an unusual RN (Webbink et al. 1987), but we now know this system actually has a black hole as the compact star. V529 Ori has a "remarkably checkered history" and it is now unclear whether the reported events are of the same star or even were nova eruptions (Webbink et al. 1987).

V1017 Sgr is an interesting case, with eruptions in 1901, 1919, 1973, and 1991, with a history and details in Webbink et al. (1987). The eruption in 1919 has a peak of seventh magnitude (or somewhat brighter) and a light curve shape that is typical for novae events. But the other three eruptions all peak at fainter than tenth magnitude and have light curve shapes typical of dwarf novae events. The small brightenings are confirmed as being dwarf nova events by their spectroscopic development (Webbink et al. 1987; Vidal & Rogers 1974). The hybrid nature of V1017 Sgr has the strong precedent from a few other hybrid systems. Most famously GK Per had a prototypical nova eruption in 1901 and has now had 14 dwarf novae events starting in 1966. So V1017 Sgr is not an RN.

As old novae (labeled as being CN) are having their second nova event discovered, the numbers of known RNe are steadily increasing. Thus, V394 CrA, V745 Sco, V3890 Sgr, CI Aql, and IM Nor have been recognized as RNe with second eruptions in 1987, 1989, 1990, 2000, and 2002. I expect that this discovery rate will speed up as more old novae are eligible to get their second eruption. These newly discovered systems usually involve stars where the first eruption is poorly observed. The known RN population is now dominated by systems with only moderate coverage and little recognition by researchers.

The currently known RNe are T Pyx, IM Nor, CI Aql, V2487 Oph, U Sco, V394 CrA, T CrB, RS Oph, V745 Sco, and V3890 Sgr. I have ordered these systems by their orbital period (as finalized later in this paper), even though the period of V2487 Oph is only approximately known. With one of the RN systems being proven only a few months ago (V2487 Oph), I realize that this list will soon have more additions.

Table 1 provides a list of these RNe, along with their primary properties. The second and third columns give the peak and quiescent brightness in the V band (V_max and V_min) as taken from Section 5. The fourth column gives the time it takes for the V-band light curve to fade by 3 mag from the peak brightness (t₃) as taken from Section 6. The next column gives the orbital period for the binary system as taken from Section 10. The final column gives the years of all known eruptions as taken from Section 3.

Several properties seen in Table 1 demonstrate a stark contrast between RNe and CNe. The RNe median amplitude is 8.6 mag (and none larger than 11.2 mag), in strong contrast with the CNe with most having amplitudes of >12 mag. The RNe are shorter in eruption duration (median t₃ of 11 days) than the CNe (median t₃ of 44 days). Almost all the CNe have orbital periods (P_orbital) of less than 0.3 days, whereas 80% of the RNe have longer orbital periods (with four of them having values of near one year). On average, the recurrent and CN populations have substantially different properties even if there is much overlap, with the RNe being generally low-amplitude, fast-declining eruptions in systems with long-orbital periods.

2.1. Subdivision of the Recurrent Novae

The RNe present us with a wide diversity of properties. It is tempting to seek divisions and categories that make sense for organizing this diversity. The divisions might be constructed for two purposes: first to identify the range of properties, and second to seek bimodality (or trimododality) which will indicate separate physical processes or histories. In the past, workers have tried to make divisions based on the orbital period (or equivalently, the luminosity class of the companion), the eruption duration (or equivalently, the t₃ value), and the light curve shape (specifically, the presence of a plateau in the tail of the eruption).

Schaefer (1990) and Schaefer & Ringwald (1995) pointed to the fact that the orbital periods come in three separate groups: very short periods with low-mass main-sequence star companions (like almost all other CVs), periods comparable to one day (with subgiant companions), and very long periods comparable to one year (with red giant companions). Such a division might be relevant for pointing to separate evolutionary paths leading to RNe, but no discussion or models have been presented to explain the three separate paths that lead to a three-fold division. A three-fold division might have consequences relating to the interaction of the nova with the circumstellar environment, for example, where a red giant companion might have a wind that can produce hard X-rays when the nova shell plows through it. But even for this, the RNe with red giants have widely varying wind densities, while there is no distinction between the two shorter period classes because both have no sensible winds. The size of the companion star will determine how much hydrogen appears in the spectrum of Type Ia supernovae, should these systems end with such an explosion. Finally, a three-fold division will produce differing spectra in quiescence, as the different luminosity classes for the companion star will be revealed by their spectra. Despite the tempting nature of this three-fold division, I do not know how to evaluate the significance of a tri-modal period distribution (see Table 1 and Figure 1), and with the small number of known RNe, I suspect that the significance is low. In principle, the size of the companion star is irrelevant to the eruption physics (which likely depends only on the white dwarf mass, the composition of the accreted material, and perhaps on the accretion rate). Most importantly, the three-fold division does not correlate with any of the eruption properties. That is, the eruption light curves, the peak absolute magnitudes, the spectral developments, the elemental abundances, the amplitudes, the accretion rates, the speed classes, the recurrence time scales, and the galactic positions all have no relation or correlation with the size of the companion star. In all, the three-fold division is relevant only for peripheral issues concerning RNe.

Zoom In Zoom Out Reset image size

Most RNe are very fast in decline, so it might be tempting to split off the relatively slow decline events (T Pyx, IM Nor, and CI Aql) into their own class (Duerbeck et al. 2002). Kato et al. (2002) suggested a variant on this by separating out just IM Nor and CI Aql into a subdivision based on similarities in the light curve and spectra. Some division based on t₃ would be relevant for the eruption physics, but I see no evidence for anything other than a continuous distribution of RNe durations. Figure 1 displays a plot of P_orbital versus t₃ for the 10 RNe, and I see no significant groupings or bimodality.

Hachisu et al. (2000a, 2002, 2008) point to a plateau in the tail of light curves as indicating an RN, with the plateau caused by an extended phase of hydrogen burning. No demographic study of nova plateaus has been made, although I will use my templates in Section 6 to quantify the situation for plateaus in RNe. The result will be that 60%–90% of the RNe are seen to have plateaus. Most critically, the RNe with plateaus (T Pyx, IM Nor, CI Aql, V2487 Oph, U Sco, and RS Oph) create a group that has no correlation with any other categorization ideas.

Kenyon (1986) has separated out T CrB and RS Oph as being in a separate class of symbiotic nova. Symbiotic stars are defined by him to be those with absorption features of a late-type giant star plus hydrogen or helium emission lines from some hot region. T CrB, RS Oph, V745 Sco, and V3890 Sgr certainly meet this definition. Nevertheless, Mikołajewska (2008) points out that these RNe are at the extreme edge of the range of properties of symbiotic stars, with short-orbital periods, high-mass white dwarfs, and very small mass ratios. Symbiotic novae are defined by Kenyon to be a very slow eruption involving thermonuclear burning of hydrogen on a symbiotic star. While the RNe might have nova eruptions on (technically) symbiotic stars, they are not symbiotic novae. Iben (2003) identifies the critical distinction between CNe and symbiotic novae as the donor star fills its Roche lobe for CNe and does not fill its Roche lobe for symbiotic novae. With this as the critical distinction, RS Oph and T CrB are not symbiotic novae (Schaefer 2009). But the exact definition is not important, because the eruptions of T CrB, RS Oph, V745 Sco, and V3890 Sgr are tremendously different from those of the symbiotic novae. Symbiotic novae have amplitudes from 2–6 mag and t₃ values from years to decades. Certainly, all symbiotic novae are fundamentally different from all RNe. That is, the fast eruptions of T CrB and RS Oph have no useful resemblance to the symbiotic novae, so any definition that lumps the two together has no utility for understanding either class.

The RNe can plausibly be divided up into groups based on amplitude (perhaps separating out IM Nor, V394 CrA, and U Sco), or recurrence time scale (separating out U Sco, RS Oph, T Pyx, and likely V2487 Oph). An extreme example of the chaos and instability in RN subdivisions is in Rosenbush (2002) wherein four categories were defined. The trouble with all these ideas is that the divisions are neither bimodal nor physically motivated, so it looks to me like all the groups are simply the arbitrary selection of some dividing line as applied to some favored property. Perhaps more importantly, these various divisions do not break up the RNe into consistent groups, which is to say that any one selected dividing line does not have any predictive power for other properties.

2.2. A New Division

What are the known eruptions from the galactic RNe? This list is important for establishing the recurrence time scale, the inter-eruption intervals, and the demographics for discovery statistics. Many of the eruptions are well known and need no particular discussion. However, eight eruptions (including one possible eruption) are little known or unpublished, and these will be discussed in detail here. Table 2 has a detailed listing of all eruptions and full information on the discovery.

3.1. Specific Eruptions

3.2. Overview of Eruptions

The photometric histories of the RNe are always determined by comparing the target system with nearby comparison stars of known magnitudes. In preparing for this paper, I have had three distinct applications which require comparison stars. First, I need accurate magnitudes in many bands for just a few relatively faint stars near to the RN to serve as comparison and check stars for differential photometry with CCD images. Second, I need magnitudes of many stars spread over the entire magnitude range of the outburst to serve as sequences for the eruption light curve on archival photographic plates. Third, I need to know the real magnitudes on a modern magnitude scale for various stars used as comparison stars by olden astronomers who have reported eruption light curve magnitudes by direct comparison with these stars.

The magnitude scale was systematized by Pogson (incidentally, the discoverer of the first known RN outburst) in the 19th century. By roughly a century ago, the standard magnitude scale was the "North Polar Sequence" defined with the Harvard plates. The sequence was extended to faint magnitudes by various methods such as comparing image diameters on double exposures where the ratio of exposure times were known precisely. We now know that reciprocity failure and other nonlinearities in photographic emulsions make for substantial systematic errors that accumulate as fainter and fainter stars are calibrated in this fashion. Magnitudes for target stars were determined by comparing star image sizes on photographs of fields double exposed on the pole region plus the target field. By the 1930s, the North Polar Sequence had been transferred to many of the Harvard–Groningen selected areas for more convenient placement on the sky. The trouble with these transfer sequences was that substantial and then-unknown errors were always present. In general, all magnitude sequences from before the 1950s were systematically distorted (from the modern scale) with the errors getting larger as the stars get fainter. Typical errors for a tenth magnitude star are from 0.5 to 1.5 mag. Sandage (2001) has made a modern study of magnitudes in the primary selected areas and finds systematic distortions of 0.6 mag at B = 8, 0.8 mag at B = 10, 1.0 mag at B = 12, and 1.3 mag at B = 14. These systematic errors in the primary standard stars were then transferred to all magnitudes in the literature. I have made detailed comparisons of many old sequences versus modern magnitudes, and I find that there is much scatter on top of this systematic problem. For comparison stars for use with Pluto, the errors were 0.7 mag in 1934 for a photographic sequence by Walter Baade and 0.01 in 1954 for a photoelectric sequence by Robert Hardie (Schaefer et al. 2008). For comparison stars for use with SN1937C, the errors ranged from 0.3 to 0.9 mag in the B band and 0.7 in the V band (Schaefer 1994). For comparison stars for use with SN1960F, the errors ranged from 0.3 to 0.9 mag in the B band and 0.7 in the V band (Schaefer 1996a). For comparison stars for use with SN1974G, the errors ranged from 0.3 to 0.9 mag in the B band and 0.7 in the V band (Schaefer 1998). For comparison stars for use with SN1981B, the errors ranged from 0.3 to 0.9 mag in the B band and 0.7 in the V band (Schaefer 1995). The point of this recitation is to emphasize that any magnitude before ∼1950 certainly has big errors while any magnitude before 1970 or so likely has big errors. Given this point, a comprehensive photometric history of the old eruptions must somehow overcome this big problem. This can be done either by remeasuring the magnitudes on the original plate material or by rescaling the old reported magnitudes based on the corrected magnitudes for the comparison stars.

The creation of modern accurate magnitude sequences had to await the widespread usage of linear detectors: the photoelectric photometer and later the CCD. Early photoelectric photometers could not go faint enough to be relevant, so it was only in the 1950s that comparison stars could be calibrated deep enough to be useful. Even then, it took many years for useful and accurate standard stars to be determined (Landolt 1973, 1983, 1992) and for RNe sequences to be measured (e.g., Kilkenny et al. 1993; Henden & Munari 2006). Many deep sequences had to await the availability of CCD cameras and the increase of telescope availability. By the 1970s or so, magnitude sequences and RNe magnitudes are all on the modern magnitude scale. For purposes of this paper, any magnitude in the literature after 1970 is taken to be correct.

The archival photographic plates (when used with modern comparison stars) are exactly on the modern B-magnitude scale. That is, the color terms are effectively zero. Another worry arises from the realization that the RNe have emission lines while the comparison stars do not, so some systematic variations could conceivably arise. But any such effects are miniscule for two reasons. First, any offset will be constant and thus not effect the shape of the light curve. For all applications of the photographic plates, we only care about the light curve shape, so an offset is irrelevant to the conclusions. This includes the comparison between photographic, photoelectric, and CCD magnitudes, with the reason being that these all have essentially constant central wavelengths and equivalent widths of the spectral sensitivity (by design). Second, the RNe have typically only 1% of their blue flux in emission lines (cf. Williams 1983). So even in the false case where slight changes in a passband allow the line light to be recorded or not, the variations will only be of order 0.01 mag. For realistic passbands with nearly constant width and roughly Gaussian shapes, the errors will always be much smaller than 0.01 mag. Thus, we can have good confidence that the old archival plate RN magnitudes have no measurable systematic bias when compared to ordinary field stars or when compared with photoelectric or CCD magnitudes.

For my three applications (see beginning of this section), I have created two sets of comparison stars for each RN. The first set is three faint stars near to the RN, which I have measured in the bands BVRIJHK. These stars are designated as "Comp" (typically a few magnitudes brighter than the normal quiescent magnitude for the RN), "Check" (typically close to the quiescent RN in both position and magnitude), and "Check2." These stars are all chosen to be well isolated from neighboring stars, and all have been found to be constant over the many observations that I have made. The observations to calibrate these comparison stars have been made many times over on many telescopes (the 0.9 m, 1.0 m, and 1.3 m telescopes at Cerro Tololo and the 0.8 m and 2.1 m at McDonald). These observations have been made with the usual full program of observations of standard stars (Landolt 1992) at high and low air mass on photometric nights. The typical measurement uncertainty in my measures is likely 0.01–0.02 mag. A number of these stars have been measured by other workers, including, for example, the T Pyx stars by A. Landolt (2005, private communication). I have JHK calibration for the stars, but the magnitudes that I will adopt are those taken from the Two Micron All Sky Survey (2MASS; Cutri et al. 2003, ¹). The magnitudes for these stars are presented in Table 3. The specific positions of these stars will be listed in Table 4.

My second set of comparison stars consists of sequences in B and V for stars spanning the range from the quiescent to peak magnitude. The reason for restricting myself to B and V for this application is that essentially all of the old literature, archival records, and photographic plates are in one of those two magnitude systems. Also, the availability of the B − V color then allows for seeking color terms in the conversion to modern magnitudes. The stars that I have selected are single, isolated, constant, and evenly spaced in magnitude as best as possible. I have also selected as many stars as possible that have been used by previous workers for their sequences. With this last point, I then have a modern and a literature value for the same sequence, and thus a reasonable conversion formula can be determined. Faint stars in the sequences are always near to the RN, and I thus have measured them as part of the calibration program described in the previous paragraph. Bright stars in the sequence are necessarily far from the RN, so I could independently calibrate only a few of them. For stars brighter than tenth magnitude or so, well-measured photometry is already available from the literature, which I have accessed and averaged from the SIMBAD database.² For many of the RNe, the AAVSO has already constructed well-measured sequences,³ often going quite deep. A variety of other sources have been used to fill in my sequences, including the Tycho2 Catalogue⁴ (Hog et al. 2000), the GSC2 Catalog⁵ (Lasker et al. 2008), and the ASAS3 Catalogue⁶ (Pojmanski 2001).

My magnitude sequences are presented in Table 4. The first and second columns are a designation for the field and reference name for the star. My star names are just sequential letters, with a few small gaps because some stars in the series were later identified as being possibly variable or otherwise unsuitable. The third column lists the positions (in J2000 coordinates) accurate to roughly 1 arcsec as useful for the unique identification of the star. The fourth column lists the formal catalog name for the star (if any). The fifth column gives the star names from the AAVSO name, which is always equal to the V-band magnitude expressed to the nearest 0.1 mag with the decimal dropped, so for example, a V = 12.34 mag star would be designated "123." The next two columns give the B and V magnitudes of the stars. The uncertainties in these magnitudes are likely to be 0.01–0.02 mag, although occasionally larger uncertainties are possible. In some cases, where the magnitude is known to be uncertain by roughly 0.1 mag or larger, the value is only quoted to the nearest tenth. The eighth column cites the source of the quoted magnitudes. The notation K1993 refers to Kilkenny et al. (1993) and H&M 2006 refers to Henden & Munari (2006). The last column gives the designations as used earlier in the literature, with a single or double letter indicating the reference followed by a colon and the designation in that reference. For T CrB, we have A (Ashbrook 1946), W (Wright 1946), B (Bertaud 1947), and Pe (Pettit 1946). For RS Oph, we have S (Shapley 1933), Pr (Prager 1940), and F (Fleming 1907). For V745 Sco, we have Pl (Plaut 1958). For V3890 Sgr, we have M (Miller 1991). Most of the lines in the comparison sequences table are only presented in the online version only.

Some of the extant magnitudes for old eruptions are now only available as specific reports in the literature. Most of these explicitly identify their comparison stars and report their adopted magnitudes for these stars. In general, these old sequences have substantial errors and these are propagated into the RN magnitudes. Fortunately, it is not too late to correct the old magnitudes. The general solution is to plot the old adopted magnitudes versus the modern magnitudes for these stars, and to fit some smooth curve so as to establish the relation between the old and modern magnitudes. With this relation, I then correct the reported RN magnitudes onto a modern magnitude scale. To this end, I have chosen many of the comparison stars in my sequences as being those relevant for correcting old sequences.

For the U Sco sequence of Pogson (Thomas 1940), we have five stars with old and new magnitudes. The rms deviation of Pogson's sequences is 0.39 mag. Nevertheless, the best-fit linear relation is that the Pogson magnitudes are close to the modern magnitudes.

Various series of magnitudes for the eruptions of T CrB have been published that need correction to the modern magnitude scale. For the 1866 eruption, we are often not told the comparison stars (nor the adopted magnitudes for the stars used), so I can only take the reported magnitudes at face value. Distortions in the light curve are likely to be minimal for times when the nova is brighter than seventh magnitude or so as the old magnitude standards were reasonable for bright stars. (Indeed, the 1866 light curve will be seen to closely match the light curve shape from the 1946 eruption down to around eighth magnitude.) Vertical compression in the light curve tail is possible, but this will be confused with the effects of the variations in the red giant companion. The comparison stars of Payne-Gaposchkin & Wright (1946) are also used by Wright (1946), with a correction of B = 0.8 × B_Wright + 1.9 for B_Wright < 11 and B = 2 × B_Wright − 11.3 for B_Wright>11. The magnitudes for the comparison stars of Pettit (1946) have a constant offset from the modern magnitudes, such that V = V_Pettit + 0.12. The comparison stars of Shapley (1933) were also used by Weber (1961) and are well represented by a constant offset with B = B_Shapley + 0.36. The comparison stars of Ashbrook (1946) have magnitudes that deviate from those in Table 4 with an rms scatter of 0.12 mag and no systematic shifts, so I will adopt his visual estimates with no corrections. Similarly, the comparison stars from Bertaud (1947), except for CrB, have an rms error of 0.11 mag with no evidence for systematic errors. The AAVSO magnitudes are also correct. The calibration of the comparison stars in Gordon & Kron (1979) is modern.

RS Oph has a variety of magnitude sequences published for the early eruption with the Harvard plates (Fleming 1907; Prager 1940; Shapley 1933). These B-band sequences have substantial differences from each other and from the modern magnitudes. For Fleming, the correction is B = B_Fleming + 1.5 for B_Fleming < 10 and B = 1.6 × B_Fleming − 4.5 for B_Fleming>10. For Prager, the correction is B = B_Prager + 0.07 for B_Prager < 10 and B = 1.4 × B_Prager − 3.93 for B_Prager>10. For Shapley, the correction is B = 1.13 × B_Shapley − 1.0 for B_Shapley < 13 and B = 0.8 × B_Shapley + 3.3 for B_Shapley>13.

Five of the dozen magnitudes for the 1949 eruption of V745 Sco are from Plaut (1958) with Leiden plates which I have not examined. Plaut explicitly states his comparison stars (although his stars h, I, and k cannot be identified with his published chart) and his adopted magnitudes; thus his RN magnitudes can be converted to a modern B-band magnitude. Plaut's magnitudes (B_Plaut) have a best-fit linear relation with my final magnitudes (see Table 4) such that B = 1.4 × B_Plaut − 5.0.

Most of the magnitudes for the 1962 eruption of V3890 Sgr are from the plates at the Maria Mitchell Observatory. These magnitudes were originally estimated by H. Dinerstein, but Miller (1991) has gone back and remeasured the magnitudes. I have not examined the Maria Mitchell plates, but we can place the magnitudes from Miller onto a modern magnitude scale by comparison of the sequences. From her comparison stars, I find that B = 1.02 × B_Miller − 0.52 for B_Miller>14 and B = 1.23 × B_Miller − 3.46 for B_Miller < 14. This correspondence will be applied to the Miller magnitudes for V3890 Sgr. Wenzel (1990) only quotes the magnitude for one of his comparison stars, but this is in the center of the range, so I adopt B = B_Wenzel + 0.24.

The light curves for the 37 eruptions have a wide variety of coverage. Four of the eruptions have only one plate to form their light curve, while many of the older eruptions have at most a few dozen plates recording the event. The more recent eruptions have extensive coverage, with the best one being the 2006 outburst of RS Oph.

Here, I collect all the existing data for most of the eruptions. (However, for a few of the eruptions with large amounts of data, my tabulations may not be complete, as some data sources may have been missed. Fortunately, in these cases, the magnitudes collected will provide a full light curve anyway.) I will present all these magnitudes explicitly in tables in this paper, because most of the magnitudes have never been published, while the published light curves are widely scattered in the literature and have been extensively revised onto a modern magnitude scale as part of this paper.

A large source of data is from the visual observations in the databases of the AAVSO, RASNZ, Variable Star Observers League of Japan (VSOLJ), and the Variable Star Network (VSNET). These observations are almost entirely in the visual band. (Almost all AAVSO work is performed with the color vision, that is with the cones in the retina of the eye for which the effective wavelength equals that of the Johnson V band, as the target is more than 0.8 mag brighter than the limiting magnitude; see Schaefer 1996b.) The modern AAVSO magnitude sequences are all accurate, so the only question is with the older AAVSO data. For estimates where the star is brighter than roughly ninth magnitude, the deviations in the comparison star magnitudes from their modern values are relatively small. For cases where it matters, I have gotten the sequences used at the time and have made the corrections if needed. The typical uncertainty in these visual magnitudes is 0.15, although these will change with circumstances and observer, from roughly 0.1 to 0.3 mag.

The AAVSO database contains ∼83% of all magnitudes of RNe in eruption. I will not include the AAVSO magnitudes in my 10 tables in the printed version of this paper, because the tabulation would take many pages of journal space and because the AAVSO provides a reliable long-term archive that is freely available to everyone. Nevertheless, the complete list of AAVSO eruption magnitudes will be presented in Table 5, with most of the 12,115 mag only visible in the online version of this paper. In this table, the first column identifies the RN, the second column gives the Julian Date of the observation, the third column gives the observing band, the fourth column gives the observed magnitude with its error bar (taken to be 0.15 mag in general), and the last column cites the source.

For many of the eruptions (also during quiescence) the primary data source is the archival photographic plates at Harvard College Observatory. This collection has roughly 500,000 plates from 1890 to 1953, with a smaller collection covering around the 1980s. Each plate is a thin glass rectangle (usually 8.5 × 10 inches in size) with a photographic emulsion on one side. Almost all plates have blue sensitive emulsion. The Harvard plates provided the original definition of the B band, and later scales for photoelectric and CCD magnitudes kept the same spectral sensitivity. So if the comparison stars that set the magnitude scale are in modern B magnitudes, then the resultant RN magnitude will also be on the modern magnitude scale. (The old magnitudes taken from the plates have been called "photographic" magnitudes, with the difference from the modern B magnitudes being only that the old comparison star magnitudes have offsets compared to modern measures; see Section 4.) Thus, with my modern comparison star sequences, the Harvard plates provide a wonderful "time machine" to get RNe light curves as far back as 1890.

The plates are taken in series, with one series for each telescope used. The plates from each series are numbered sequentially, with a capital letter series identifier followed by the sequential number. For example, one of the deeper plate series (taken with the 24 inch Bruce Doublet telescope) is designated with the letter "A," such that plate A1 was taken in 1893 and the last A plate (A27504) was taken in 1950. The better A plates have a limiting magnitude of roughly 18, with of order 30 useable plates that show the field of any one RNe. The primary deep series (with limits of 16–18 mag) are the A, MC, and MF, while there are many tens of thousands of plates with less deep limits (with limits of 13–16 mag) in the B and I series. Harvard also has several series (chiefly RH, RB, AM, AC, and the Damons) that are patrol plates, where large areas of the sky are regularly photographed to moderate limits (typically 12–15 mag). The entire sky is covered roughly equally, with typically 1000–3000 plates covering any particular RN. Further details are available in Collazzi et al. (2009) or at the Harvard archive Web site.⁷

The 1σ uncertainty in the magnitude depends substantially on the plate material, with various field effects increasing the uncertainty toward the edge of the plate. For a well-exposed field with a good sequence of comparison stars, measurements with iris diaphragm photometers can give a median accuracy of 0.08 mag, and can get as good as 0.04 mag (Schaefer et al. 2008). However, most of the plates were measured by direct visual comparison of image sizes by means of a view through a loupe (a handheld magnifier). I have a very large amount of experience in this, and I also have performed a variety of experiments over the years for the accuracy of these comparisons. The median accuracy is 0.15 mag, with typical accuracy ranging from 0.1 to 0.2 mag. It is difficult to evaluate the uncertainty for any one individual plate, so I have used a default 1σ error of 0.15 mag unless my notes point to some other value as quoted in the tables. While the measurement uncertainty for plates is roughly a factor of 10 worse than usually obtainable with photoelectric or CCD techniques, this uncertainty is comparable to or smaller than the usual random flickering. Thus, for many purposes, the moderate accuracy of photographic magnitudes is perfectly fine and just as good as more "modern" measures.

The eruption light curves are collected together into one table for each RN (Tables 6–15). The format for each line will come in one of the two forms, depending on whether the B-band and V-band magnitudes were made effectively simultaneously. For most observations, the columns will list the Julian date of the observation, the band, the observed magnitude (along with its uncertainty), and the source of the data. If B and V magnitudes are from the same time, then the second and third columns will instead report the B-band and V-band magnitudes in that order. (This is simply to make the tables more compact.) Each eruption will have a header specifying the year of the peak. The source of the data might be the Harvard plate number (e.g., B5289, AM1172), the organization reporting the data, or the literature reference. The observations are listed in time order. The complete eruption light curves consist of those magnitudes from each of the tables for individual RNe (Tables 6–15) plus the AAVSO magnitudes for many individual eruptions collected into Table 5 (most of which is only visible in the online table). For the next 10 paragraphs in this section, I will describe the data sources specific to each of the 10 galactic RNe.