The Pisgah Automated Survey: A Photometric Search for Low‐Mass Detached Eclipsing Binaries and Other Variable Stars

The probability of detecting low‐mass detached eclipsing binaries depends on the intrinsic characteristics of those systems, such as their space density, their number distribution with mass (initial mass function), their apparent magnitudes, the ratios of sizes and temperatures of their components, their spatial inclinations, and the distribution of their orbital periods. The combination of the last four parameters determines the width and depth of the eclipses our survey seeks to detect. To be successful, our equipment must conduct adequate time sampling of stars in a sufficiently large volume of space to make detection of new binaries a certainty. This requires an intelligent trade‐off between areal coverage of the sky, limiting magnitude for eclipse detection, and sampling rates.

To investigate these trades, we have developed a set of Monte Carlo simulations that incorporate our best knowledge of low‐mass binary properties to estimate the number of low‐mass eclipsing binaries detected by a given instrumental setup.²

The simulations generate a sample of stars with the space density and mass distribution derived by Reid & Gizis (1997) using stars within 8 pc. The program then assigns binary companions to 35% of those stars, where the companion mass is drawn according to the binary mass‐ratio distribution observed by Reid (1997); no triple systems, or higher, are included in our sample. In the absence of a well‐measured period distribution for binaries below 1 M_⊙, we assign orbital periods to the binaries using the distribution found by Duquennoy & Mayor (1991) for G stars. The simulation then distributes the binaries in space with random inclinations and with a space density that drops above and below the Galactic disk, following a two‐component exponential to account for the thin and thick disk components (see Gilmore, King, & van der Kruit 1990). The binary magnitudes are assigned by look‐up from the mass‐magnitude relations of Henry & McCarthy (1993), and the radii are calculated from the approximate mass‐radius relation of Neece (1984), which we chose to minimize computation time.

Once the sample is generated, the program selects those binaries that will show eclipses based on their orbital inclinations, orbital periods, and the radii of the stars. The simulation discards interacting binaries, defined as those in which at least one component fills its Roche lobe. It also eliminates all the binaries with eclipses narrower than 5% of the orbital period, because detecting these would require over 300 observations, a number we decided a priori to be impractically large.³ The sample of synthetic binaries that remain represents an estimate of all the binaries that are detectable, in principle. Specific choices of hardware will reduce this number, so the final step is to impose detection limits dictated by various selections of hardware and observing strategy. Note that we have not placed any upper limit on the orbital periods; instead these limits occur naturally, since the eclipse probability falls so rapidly with increasing period that practically no binaries with periods longer than about 10 days are expected.

The results of the simulations are presented in Figures 2 and 3. Figure 2 shows an I versus V−I color‐magnitude diagram containing the 131,348 detectable low‐mass detached eclipsing binaries predicted by the simulation in a search area covering the entire sky (this diagram is instrument independent). The vertical cutoffs in this diagram at V− I∼4.2 and 1.37 result from restricting the masses in the simulations to between 0.82 and 0.1 M_⊙. The cutoff at the top of the diagram reflects the declining volume of space with decreasing distance to the observer. The number of binaries increases with increasing magnitude in a manner that reflects the competition between increasing search volume and the exponential decline in stellar space density associated with the two components of the Galactic disk. The number of detectable binaries decreases toward redder colors, in spite of an increase in the mass function, because the eclipses grow increasingly narrow as the stellar radii decrease. It is likely that this trend underestimates the real number of low‐mass systems, because the binary period distribution we used in our calculations decreases monotonically at short periods, while observations of nearby low‐mass binaries (Reid & Gizis 1997) and those in the Hyades (Duquennoy & Mayor 1991) show suggestions of an increase in numbers at short periods. Even a modest increase in the number of short‐period binaries can significantly change the number of detectable binaries at the low‐mass end. The two bands with slightly higher concentrations of binaries, located between 1.4≲V− I≲1.9 and 2.1≲V− I≲2.7, reflect changes in the slopes of the magnitude transformation equations from Reid & Gizis (1997) and the mass‐M_V relation in Henry & McCarthy (1993). Both slope changes cause a large range of masses to accumulate in a narrow range of colors.

Zoom In Zoom Out Reset image size

Zoom In Zoom Out Reset image size

Superimposed in this diagram we show two horizontal dotted lines at I = 12.0 and 19.5, which represent the effective detection limits of two published photometric surveys we have used to calibrate our simulation.⁴ Those surveys, both conducted by the Warsaw University Observatory in Poland, are the Optical Gravitational Lensing Experiment (OGLE; Udalski et al. 1992) and the All Sky Automated Survey (ASAS; Pojmanski 1997). Figure 3 shows the number distributions of binaries predicted by our simulations for those two surveys after applying the corresponding survey‐dependent completion functions. These functions are necessary to model the dependence of detection on magnitude, since shallow eclipses can be detected in bright stars but not in faint ones.⁵ The decrease in numbers below V− I∼1.4 is artificially steep in all the plots because of the upper mass limit we imposed.

In actuality, the OGLE experiment detected two foreground eclipsing M dwarfs in a 2.4 deg² field centered at the Small Magellanic Cloud (Udalski et al. 1998), while ASAS found two new binaries with estimated masses of about 0.8 M_⊙ in its first phase covering 300 deg² (Pojmanski 2000). The triangles in Figure 2 show the estimated locations of these four systems. The ASAS detections are at the extreme upper end of the mass range we considered, so we cannot reliably compare them to the expected rate. For the OGLE survey, the real detection rate is smaller by a factor of 2.5 than the expectation from our simulations, which may be only a statistical variation or simply indicate that the OGLE survey incompleteness does not follow the functions we applied to the data. Both OGLE detections were 1.0–1.5 mags brighter than the magnitude limit, favoring this possibility. Alternatively, the value we have used for the space density of low‐mass stars, 0.06 stars pc⁻³ (Reid & Gizis 1997), may be too high. The number is based on known stars within 8 pc and could be low if the sample is incomplete, but it is difficult to argue that it is too high. It is even more unlikely that our simulation overestimates the number of short‐period binaries, since the 8 pc sample actually contains three such eclipsing binaries, while our simulation predicts none.⁶ Whatever the explanation for the shortage of detections, our search strategy will be similar to that of the OGLE survey, so we should have a similar ratio between expectation and detection, and can scale the output of our simulation accordingly.

Figures 2 and 3 reveal that a survey with the characteristics of ASAS will not likely detect M dwarf eclipsing binaries, since the number of detectable systems in the whole sky drops to almost zero at V− I = 2.0. OGLE, on the other hand, has no difficulty detecting M dwarf systems, but has only small areal coverage. Consequently, OGLE will mostly detect binaries too faint for practical spectroscopic follow‐up.

We would like our survey limit to be bright enough that all the binaries detected will be candidates for spectroscopic follow‐up on 4 m class telescopes or smaller. At the same time we need to reach deep enough in magnitude to successfully detect M dwarf binaries in a reasonable amount of sky. Our simulations show that a binary detection limit of I∼13.5 is a reasonable value to meet those two requirements; below this magnitude the S/N on readily available high‐resolution spectrographs becomes too poor to accurately measure the binary’s radial velocity curves. Based on the ASAS survey, in which the faintest variable is 1.0–1.5 mag brighter than the survey’s photometric detection limit, we estimate that detecting binaries with magnitudes of 13.5 will require a photometric detection limit of I∼14.5–15.0.

The solid horizontal line in Figure 2 shows the results of the simulation for a binary detection limit of I = 13.5. With this limit we can detect systems down to V− I∼3.0, which corresponds to primary masses of about 0.2 M_⊙. The histogram of color distribution for all the binaries above this limit is shown in the central plot of Figure 3. There are about 3122 detectable binaries in the whole sky, and 202 binaries with V− I≥2.0 in this histogram. Thus, we expect that a survey with a detection limit of I = 13.5 will find one low‐mass detached eclipsing binary in every 13 deg² surveyed, and one M dwarf binary in every 204 deg² of sky surveyed. If we adjust the detection rate downward based on OGLE detections, these numbers become 325 and 510°, respectively. This is an acceptable yield for a dedicated telescope, and we shall see in § 3 that the requirement of 15th magnitude photometry is easy to satisfy using equipment from the high‐end consumer market for astronomical hardware.

The telescope‐camera combination was selected as a trade‐off between aperture, sky coverage, and pixel scale. We needed the largest practical field of view that a commercially available CCD camera could cover. This translated into a need for short focal length, which we met by choosing an 8 inch (203.2 mm) Meade telescope with an optional f/6.3 focal ratio. The usable field of view of this telescope, without vignetting, is 037 × 037, but we are using the entire field provided by the CCD (≃ 128 × 128), which, in spite of the vignetting and coma aberration in the images, still provides acceptable photometry. The coma results in an increasing radial elongation of the shape of the stars as we move away from the center of the images. This aberration, combined with the vignetting, results in a progressive decay of the S/N of the stars toward the edge of the images. These effects are illustrated in Figure 4, where we show how the shape and the S/N of two stars of similar magnitude vary, depending on the location of the stars in the images.

Zoom In Zoom Out Reset image size

The CCD is the largest available from Apogee Instruments, Inc., a 28.7 × 28.7 mm chip with 0.014 μm pixels. On the Meade telescope, this yields a pixel scale of 2256 pixel⁻¹. In practice, the seeing at our site is frequently so bad that we bin 2 × 2 to reduce the readout time. Thus, an after‐the‐fact analysis shows that we could have saved about $6000 by choosing a 1024 × 1024 CCD with 24 μm pixels.

To reach our magnitude limit (I∼15) with a S/N of 3 requires 3 minute integrations.⁷ At that integration time, stars brighter than I≃8.5–9.0 saturate. A larger telescope with smaller field would accomplish this faster, but CCD readout (10–15 s in the nonbinned configuration) and slewing between fields (≃20 s) would diminish the duty cycle, resulting in little gain for the extra cost. In 3 minute exposures, the noise from sky brightness in the I‐band dominates the dark noise from the camera, which is two‐stage thermoelectrically cooled to - 25 ± 1–2° C. The CCD is a front‐illuminated Thomson THX7899M device. Its dark current at operating temperature is ∼0.16 ADU s⁻¹. The CCD gain is ∼14 e^- ADU⁻¹, and the readout noise is ∼30 e^-. In the 2 × 2 binned mode, readout requires 5–6 s.

The telescope was delivered with a Meade LX‐200 computerized mount, which we used to begin operations. We quickly realized that this mount was inadequate in two respects. The pointing accuracy, even after applying careful corrections, was no better than 8–10, and tracking trails 10–15 long appeared after 3–4 minute exposures. These trails and pointing errors made the automated astrometry (see § 4) very difficult and time consuming. In addition, the mount began to experience random runaways in right ascension (R.A.) after a few month’s operations. We decided to replace this mount with a much better quality (and more expensive) one, the German‐equatorial GT‐1100S Paramount from Software Bisque. The pointing accuracy with this new mount is better than 30–45 after correcting systematic pointing errors using a TPoint model.⁸ The slew rate of the Paramount (5° s⁻¹) and the large field of view of our telescope also make the system ideal for rapid response to targets of opportunity, such as gamma‐ray bursts.⁹

The camera is mounted to the telescope using a custom‐made connector that includes room for a single 50 mm filter. Our intended targets are very red, so we have inserted a Bessell I‐band filter with a central wavelength of λ_center = 8000 Å, a bandwidth Δλ_FWHM≃1500 Å, and a peak transmission of ≃97%. Figures 5 and 6 show the final instrumental setup. Total hardware costs, including the control software, appear in Table 2.

Zoom In Zoom Out Reset image size

Zoom In Zoom Out Reset image size

The telescope and the CCD are controlled by programs from Software Bisque called TheSky and CCDSoft, which are in common use among amateur astronomers. We have wrapped these commercial products inside a set of custom observing routines that coordinate the operations of all the hardware, including the dome. The result is a fully automated system for data collection. We describe the software in more detail in § 4.1.

The selection of the site to install the survey’s equipment was a compromise between sky quality¹⁰pasp_116_815_22tb3 and proximity to Chapel Hill. Since this was to be a prototype and would likely require a lot of troubleshooting, we decided to accept an offer from the Pisgah Astronomical Research Institute (PARI)¹¹ to install the hardware at their facilities in Rosman, North Carolina, 270 miles from our home institution. PARI also provided site development and constructed the enclosure at their expense.

PARI personnel constructed a building to house the telescope at the highest point on the site, at an elevation of about 1800 feet (548 m) above sea level. The building consists of a two‐story structure that includes battery backup power, air and humidity evacuation systems, a control computer, and fiber optics connections to allow remote access to the equipment. The structure is topped by a 10 foot (3.05 m) diameter dome acquired from Technical Innovations, Inc. The dome includes the Digital Dome Works software package, which provides the necessary tools to control the dome, and a small weather station to ensure automatic closure in the event of adverse weather conditions. Besides the weather, which prevents observation 40% of the time, the dome has been responsible for most of the lost of observing time, suffering a variety of mechanical and electrical failures. Future projects at PARI will use roll‐off roofs for improved simplicity and reliability.

Survey operations and subsequent data analysis are handled by three fully dedicated computers. The first one, located in the first floor of the survey’s building at PARI, is a Pentium III, 550 MHz PC that runs the survey software and controls the telescope, camera, and dome. It also serves as temporary storage for the data collected each night. The second computer is a Pentium II laptop, which provides remote access to the survey at any time and from any location, using an ethernet or telephone connection and the software package PCAnywhere from Symantec Corporation. Ethernet access to PARI is via a T1 line, which allows us to transfer the approximately 0.4 Gbytes of data collected each night at a speed of ∼0.24 Mbyte s⁻¹. The third computer is a Dell Precision Workstation 530, which is used to reduce and analyze the data.

In addition to these three computers, the survey accesses two other machines in the PARI local network, which respectively display the local weather conditions through a set of TV cameras placed around the site, and information from an on‐site GPS receiver. The internal clock on the survey computer synchronizes itself to the GPS every 20 minutes, to assure precise tracking of time in our time series photometry.

The most difficult cost to contain in an automated telescope project is software development. It is in this category that we have realized the greatest savings by exploiting consumer‐level software. Each of the commercial software packages used to operate the telescope, the CCD, and the dome include libraries of ActiveX functions developed by the Astronomy Common Object Model Initiative¹² (ASCOM). Those functions can be called from Visual Basic or other Microsoft languages. We have written a Visual Basic script that manages the automation of the telescope, dome, and CCD through these library routines. A flowchart of the program is shown in Figure 7.¹³ We are currently in the process of porting the code to C# in the .NET framework, which will allow automatic response to Web‐generated interrupts for observing gamma‐ray bursts, afterglows, and other targets of opportunity.

Zoom In Zoom Out Reset image size

The program, called SurveyLoop.vbs, runs continuously as a background process in the computer installed at the dome. It keeps track of the time of the day through the computer’s internal clock. Two hours before sunset, it begins to cool the CCD to the preset temperature of - 25° C. At sunset, after checking the weather conditions, the program opens the dome and slews the telescope to a position close to zenith, where it begins taking a series of sky flats, followed by dark frames. Once it finishes taking the darks, the program waits until the end of the astronomical twilight to begin monitoring the list of preset target fields. The integration time for all the fields is 180 s, which yields a S/N≃3 for I = 15.0.

Initially, our list of targets were located along the celestial equator, to allow follow‐up observations from either hemisphere. The telescope was repointed in R.A. after each exposure, monitoring those fields on the list of targets that fell within ±2 hr of h.a. = 0. Using this strategy we were able to repeatedly image 60–70 different fields per night, covering a sky area of 107.0–115.3 deg². During the first year of operation, we found two major problems with this scheme. First, the fields were imaged only four times per night, and we discovered through experience that this made it impossible to collect enough fields for binary detection in a single season. Second, the frequent long slews in R.A. required the already problematic dome to move quite often, leading to frequent misalignment problems between the shutter and the telescope, and to mechanical failures. To avoid those two problems, we changed the spatial distribution of target fields to sets of 10 (2 × 5) adjacent fields in a rectangular area of ∼16.5 deg² (257 × 642). The survey then cycles over the fields in one set while they are within hour angles of ±3 hr, then moves to the next set of fields, strategically located 5–6 hr later in right ascension. Each set of fields is monitored until ∼300 clear images are available, or until it falls out of the h.a. coverage.¹⁴ We avoid fields in the plane of the Galaxy, because they are too crowded for automated reduction to work properly (see § 4.2.1). This is not a significant loss, because our survey will not cover the whole sky, and the distribution of our target stars is essentially isotropic (our magnitude limit defines a volume that does not extend out of the Galactic disk for M dwarfs). Our new strategy has reduced the number of dome failures dramatically, and it allows us to acquire at least 8–10 images of each target field per night, with a cycle time of about 35 minutes. So it is in principle possible to collect enough frames for binary detection in one observing season.

Between each exposure, the program checks the weather conditions and the time. If either an adverse weather or morning twilight notice occurs, the equipment is shut down following one of the two procedures outlined in Figure 7. If the notice indicates adverse weather conditions, the program closes the dome, parks the telescope, and remains idle, ready to resume observations if the weather conditions improve. If the notice reports the beginning of the morning astronomical twilight, the program parks the telescope, takes a second set of dark frames, and turns off the CCD cooler. It then goes back to the beginning of the loop, where it waits until the next sunset.

The survey collects a total of 130–200 images per night, including flats, darks, and target fields. These are analyzed off‐line by the reduction and analysis protocol outlined in Figure 8. The reductions are done at the end of each night, but the analysis to find eclipsing variable candidates must include more than a single night’s data. Our current experience shows that at least 10 observations during eclipse are required to trigger our period search routine (see 4.2.2). If we assume that the observations occur at random orbital phase, then 300 observations are required to ensure a greater than 95% detection probability for binaries with eclipses equal to 5% of their orbital period. The exact number of observations depends on the ratio of the stars’ radii, which we do not know a priori, and on the photometric precision of the light curves. For binaries with longer eclipses, the number of observations needed is smaller than for those with shorter eclipses. In addition, the number of observations necessary to detect the eclipses is larger for fainter stars, as the photometric dispersion of the light curves increases with magnitude. As a result, the survey will be increasingly incomplete, because we remove fields from the target list after 300 frames are collected.¹⁵ A full analysis following the scheme in Figure 8 includes all the images collected for each field, which usually take several months to acquire.

Zoom In Zoom Out Reset image size

Our technique for finding variables relies on having “light curves” of each star available, so the analysis must automatically identify stars based on their positions and assemble photometric measurements from multiple images into a time series. Only then can we look for the periodic brightness changes that will signal the discovery of an eclipsing binary, or any other kind of periodic variable.

In our reduction and analysis routines, we have again saved on software development by relying heavily on available code, mostly within the Image Reduction and Analysis Facility (IRAF) packages from NOAO. Our protocol consists of three main programs: generaredroutine.C, generaanalroutine.C, and periodsearch.C. These programs generate IRAF scripts called reduceall.cl, analyzeall.cl, and searchperiod.cl. These scripts then direct a fully automated analysis of the data within IRAF. The final output is a set of folded light curves of all the variable star candidates detected in the monitored fields. In the following subsections, we describe each of the scripts in more detail.

4.2.1. reduceall.cl

4.2.2. analyzeall.cl and searchperiod.cl

The outputs from reduceall.cl are the inputs to the next script, analyzeall.cl, which derives astrometric solutions for each image and cross‐identifies each star in all the images of a given field so light curves can be generated. The script uses the package WCSTools (Mink 1997, 2001) to perform astrometric solutions for each target field, using list of stars from the corresponding images of the digitized sky survey POSS2 as references. After finishing the astrometry, the script can make light curves from the dates and times in the image headers, and the magnitudes from the photometry files, but the latter must first be corrected for extinction and calibrated against a standard system.

To accomplish this, we select a reference frame for each target field from the night with the best photometric quality. The selection of the reference frames needs to be done by hand, although only once per target field. For the reference frames only, we conduct calibrated photometry to produce a single reference image of each survey field. The raw magnitudes in that frame are transformed to the Johnson‐Cousin system by using the magnitude calibration technique developed by Harris, Fitzgerald, & Reed (1981). We use l = 3–4 images of each target field from the selected night, with air masses χ∈[1,2]. In each image we select as standards m = 10–12 isolated stars uniformly distributed across the frame (avoiding the edges). The selected objects have no saturated pixels and have magnitudes published in the Tycho‐2 Catalog (Høg et al. 2000), with Tycho B_T−V_T colors in the range - 0.2≤B_T− V_T≤2.0. The Tycho standard magnitudes of those stars are converted to I‐band Johnson‐Cousin magnitudes using the transformation equation derived by Richmond et al. (2000). Finally, the raw magnitudes are transformed using the standard transformation equation

This equation does not include color terms, because we monitor the stars in only one filter, so no color information is available. The calculated values of the a₁ and a₂ coefficients are a₁ = 9.509 ± 0.097 and a₂ = 0.051 ± 0.070. The uncertainties in the estimated I_std magnitudes are of the order of σ_I≃0.13 mag.

The raw magnitudes of all the other images of each field are corrected to the level of the reference frame by calculating and applying a single magnitude correction ΔI. This correction is an average of the difference measured for each star in the frame. Magnitude difference diagrams for some sample images are represented in Figure 9 and show that there is no systematic bias with brightness in the photometry. We discard images with ΔI≥0.5, because their photometric quality is too poor.

Zoom In Zoom Out Reset image size

Once the magnitudes are adjusted in this way, the script assembles a separate light curve for each star, including now all the data collected up to this time. It is possible to identify variables from their light curves at this time using the strategy employed by Pojmanski (1997), who selected stars with standard deviations σ_I larger than expected for their I magnitude. Figure 10 shows graphically how this strategy would work for one of the fields monitored by the survey (stars above the dashed line in the figure are variable candidates). However, this strategy is not optimal for our survey, because it does not use all the available information. We know that the eclipsing binaries we seek are periodic, so we can implement more sensitive techniques by searching for periodic variations. We have chosen to use the technique of “analysis of variance” (AoV; Schwarzenberg‐Czerny 1989), which was also employed as a period search method by ASAS (Pojmanski 1997) and which excels at detecting narrow periodic features.

Zoom In Zoom Out Reset image size

The plot in Figure 10 is from a field that includes one of the three known eclipsing M dwarf binaries, GJ 2069A. As we discuss in § 5, this binary has eclipses only ∼0.2 mag in depth. Its location in Figure 10 is marked with a dark square and does not clearly stand out compared to the other variable star candidates. Many of the stars that fall above the curve are false positives caused by blended stellar profiles. If we apply a selection threshold capable of extracting GJ 2069A from this diagram, it will generate too many false candidates to be useful. On the other hand, the AoV method finds GJ 2069A easily by looking for variance in its folded light curves, as demonstrated in § 5.1.

The last script in our analysis performs the AoV method on all of the light curves generated for each field observed. We have used the public aov.f routine written by Schwarzenberg‐Czerny (1989), but we call it from IRAF for convenience in automation. The program applies a one‐way AoV algorithm, which Schwarzenberg‐Czerny (1989) has shown is much better than the Fourier transforms at detecting periodic narrow pulses. After folding the light curves over a specified range of trial frequencies, the program generates lists of values of the standard AoV test statistics, Θ_AoV, for those frequencies (we call plots of Θ_AoV versus frequency “AoV periodograms”). Any periodic feature in the light curves appears in those AoV periodograms as a high peak at the corresponding frequency. We refer the reader to the original paper by Schwarzenberg‐Czerny for a more detailed description of this technique. For our purposes, we have restricted the range of trial frequencies to 0.1–10 day⁻¹, since, as indicated in § 2.1, the probability of detecting binaries with orbital periods larger than 10 days is close to zero.

To identify candidates from the output of the AoV routine, we define a “contrast parameter” by calculating the ratio of the largest peak above 5 day⁻¹ and the largest below 5 day⁻¹. When that ratio is less than a given threshold value, the star is considered a variable candidate. The value of the threshold was calibrated using the AoV periodogram of a variety of variables, the M dwarf binary GJ 2069A among them (see § 5.1). This new identification technique eliminates the problem of false detections caused by blends, which do not show, in general, any detectable periodic behavior.

Finally, the script automatically generates, for all the possible variable stars, supermongo (Lupton & Monger 1998) plots of their light curves folded at the periods identified in the AoV routine. The folded light curves must be inspected by eye to make a final determination about the existence and nature of the variability.

In order to test the performance of our variable identification technique, we included the previously known M dwarf binary GJ 2069A among the first set of stars monitored by the survey. This binary in fact provides an excellent test case, given that its characteristics are far from optimal for a successful detection in our images, as described below. Admirably, our routines have no problem identifying the object as a variable candidate (see Fig. 15), although some problems were found when we tried to recover its orbital period from the folded light curves.

The orbital inclination of GJ 2069A is very high, which translates into shallow eclipses of only 0.2 and 0.15 mag in depth (Delfosse et al. 1999). These values are close to the dispersion limit of our photometry for the apparent magnitude of the binary. Therefore, its light curve is at the marginal limit of the survey for a successful identification by visual inspection. Furthermore, given the low resolution of our images, the binary appears blended with a brighter nearby star, as shown in Figure 13. The magnitude of the blend is I∼9.02. As we mention in § 3, stars of that magnitude saturate in some of the images, resulting in a larger magnitude dispersion in their light curves. In addition, both stars in the blend are M dwarfs and therefore magnetically active, with sporadic flares and continuous surface brightness variations. This larger magnitude dispersion, combined with the shallowness of the eclipses, inhibits a clear identification of the binary from its folded light curves (notice that we list it in Table 5 as a variable of type “unknown,” because we could not recognize its Algol type light curve in our visual inspections). Figure 14 (top) shows the current light curve of the blend in MHJD, and (bottom) after folding it using the known orbital period of 2.771468 days, derived by Delfosse et al. (1999). The secondary eclipse (at phase ∼0.5) is obvious in the folded light curve, but only because we know its location a priori. There are no data during primary eclipse, which would occur around phase = 0 in that plot. The magnitude dispersion of the baseline is significant because of the combination of effects mentioned above.

Zoom In Zoom Out Reset image size

Zoom In Zoom Out Reset image size

With more data covering all the orbital phases, even a binary in such marginal conditions in our survey as GJ 2069A can be eventually identified by eye. But with the current light curve coverage, the binary is already clearly identified as a variable star by the AoV periodograms, as illustrated in Figure 15. This figure (top) shows the combined AoV periodogram of GJ 2069A and its blended companion, compared to (bottom) the AoV periodogram of a nonvariable star. While the periodogram of the nonvariable is basically featureless, several variance peaks are clearly visible in the one for GJ 2069A. In particular, the peak at a frequency of about 0.361 day⁻¹ in the periodogram reproduces closely the known orbital period of P∼2.771468 days for the binary, proving the viability of our variable candidate selection algorithms.

Zoom In Zoom Out Reset image size

The identification of GJ 2069A as a periodic variable demonstrates that the Pisgah Survey’s search strategy can successfully detect M dwarf eclipsing binaries. However, our less successful identification of its period or light curve type shows that it will not always be possible to go from detection to follow‐up observations without an intermediate step. Better seeing conditions and a slightly higher pixel resolution can reduce the number of cases in which blends of this type become a problem. Also, continued Pisgah Survey measurements of fields with periodic variables of unknown type can ultimately resolve their light curves as more cycles are measured, but this is not the most efficient use of survey time. More practical, in this case, would be follow‐up photometry from other automated telescopes, such as those designed for gamma‐ray burst optical transient observations, which usually have ample time between bursts to conduct synoptic observations.

Even without intermediate observations, two‐thirds of the periodic variables that we detected have identifiable light curves and can be followed up immediately. At this point, none of these are the low‐mass eclipsing binaries we seek, but this is consistent with our expected detection rate of one every ∼30 deg².

We would like to thank University of North Carolina alumnus Dr. Henry Cox for his private donations to this project and for suggesting the name “Pisgah Survey.” We also thank Don Cline, president of PARI, for making the institute’s facilities fully available to us and for his contributions to the design and construction of the survey’s building. We are specially grateful to all the PARI personnel for the incalculable amount of help that we have received from them on the setup and maintenance of our equipment. Special thanks to Dr. Eileen Friel, of the National Science Foundation, for revisions of this paper and for her many constructive comments toward its completion; and to the referee, Dr. Michael Richmond, for his helpful comments and suggestions. This project has been partially supported by the National Science Foundation through grant AST 00‐94289.