JSTOR

1. INTRODUCTION

High‐precision, wide‐field, time‐sampled stellar photometry has a variety of applications, e.g., searches for transits by extrasolar planets, the study of microlensing events, low‐level stellar variability, and searches for nearby, low‐mass binaries. In each case, the stellar populations, time sampling rates, photometric precisions, and data reduction requirements will be different. Since achieving the highest precision theoretically possible is often necessary to meet certain science requirements, it is important to determine what natural limits exist to photometric precision and what can be done to optimize the observations and data reduction procedures. Here we are particularly interested in obtaining high photometric precisions, e.g., a few millimagnitudes or better, for long‐term stellar light curves. This is the requirement for surveys to detect transits by extrasolar planets and to understand long‐term, low‐level intrinsic stellar variability. We show that it is possible to obtain long‐term light curves at these high precisions over a wide field of view and discuss the differential (aperture) photometry techniques we use.

We employ a time series of V‐band exposures toward a 1 deg² field, sampled at 5.7 minute intervals over the course of 5 nights. Our primary interest in obtaining these data was a wide‐field survey for the transits of extrasolar planets. We perform aperture photometry on each star in the field, yielding high‐precision differential light curves. In § 2 we describe our observational program using the MOSAIC CCD Camera and 0.9 m telescope at KPNO. In § 3 we describe our detailed data reduction. In § 4 we present the results of our survey and discuss these results in § 5 with particular attention to the detection of transits by extrasolar planets.

2. OBSERVATIONS

We observed two different fields centered at ^h06^m00^s, ( , ) and ^h09^m30^s, ( , ) on the nights of 2000 March 16–20 UT using the KPNO MOSAIC Camera at the f/7.5 0.9 m telescope. The eight pixel², thinned, SITe CCDs of the MOSAIC Camera cover an area of with a plate scale of 0 43 pixel⁻¹. KPNO documentation reports that the CCDs are linear to 0.1% up to 70,000 electrons with an average gain for the CCDs of 3 e⁻ ADU⁻¹. Our typical seeing of 1 4 FWHM resulted in an effective magnitude range for our survey of . The images were guided using the facility Leaky guider. This guider was effective at keeping stars within a few pixels of their nominal positions and allowed us to realign the telescope to the same fields each night with an accuracy of ∼1 pixel.

The 5 nights of observations consisted of a time series of exposures of the two fields in the V band with exposure times of 3 minutes, sampled every 5.7 minutes, with this duty cycle imposed by the CCD readout time. Each field was also observed once or twice in UBRI filters with exposure lengths set to ensure that most stars that are detected in V also have detections at reasonably high signal‐to‐noise ratio in these other passbands. A field of standard stars from Landolt (1992) was observed in UBVRI in conjunction with the observations of the survey fields in order to calibrate the magnitudes in each filter. The UBVRI colors are used to estimate the spectral type and luminosity class of each star.

Flat fields were obtained for each filter. For the V‐band filter a large number of flats (120) was obtained over the 5 days of the observing run and was later reduced to produce a single flat‐field calibration image. The same reduced V‐band flat was used to reduce each night’s data. Bias frames were also taken throughout the run to combine into a single frame for all reductions.

Observing conditions varied throughout the run. During most of the observations, thin cirrus clouds were present, and since it was bright time, the sky background was relatively high. This was not generally a problem because our main concern was differential photometry of the brightest stars in the field for which we could obtain the necessary high‐precision light curves. The signal‐to‐noise ratio for measuring these stars is dominated by the Poisson fluctuations from the star itself rather than the background. Of greater concern were the fluctuations in seeing. The seeing varied from 1 1 to 2 5 FWHM over the course of the run, changing with both weather conditions and air mass.

3. DATA REDUCTION

Each CCD image is initially reduced by subtracting a bias measured from the overscan region, then further corrected by subtracting a residual two‐dimensional bias pattern that is calibrated using an averaged set of bias images. V‐band dome flat fields, which have been debiased in this method, are averaged into one frame for each night of the observations after rejecting cosmic rays from each exposure. Although there were large‐scale flat‐field differences of about 1% between flats observed on different nights, probably due to differences in the alignment and illuminations of the flat‐field screen and/or filter, these are ignored because their effect on the internal photometric precision was negligible. We are not concerned with calibrating absolute magnitudes to this (1%) level of accuracy. A master V flat was produced for the entire run by taking a mean of the nightly flat fields. The master V flat combines 120 individual exposures of 40,000 e⁻ pixel⁻¹ apiece to obtain a fractional uncertainty of ∼ (per pixel) in our flat, so flat‐fielding is not a limiting factor in our final photometric precision (see, e.g., Newberry 1991). The UBRI flats are filtered averages of typically 20 dome‐flat exposures. The flat fields for each CCD are normalized to a mean value of 1 by multiplying each by a unique normalizing factor.

We use IRAF scripts (Tody 1986) to locate objects, measure instrumental fluxes in circular apertures, and record this information in files for each exposure of the V‐band time series. First, a group of images with good seeing is chosen, registered, and averaged. We use this master image to locate, using daofind, all of the stars in the field bright enough that they would be detected in individual exposures. Next, the shifts between all images of the field are found by correlating these images to the master image. These shifts are used to create a list of the expected coordinates for the stars in each image. Finally, the IRAF aperture photometry task PHOT is used to measure the flux of each star and to report on errors encountered while performing the aperture photometry. A file is written for each exposure and CCD listing the time of observation, location of each star in the image, its flux, the local sky flux, the star’s instrumental magnitude and magnitude error, and IRAF’s own error flags. Each star is measured in five different aperture sizes having radii of 3, 4, 5, 6.5, and 7.5 pixels (1 3, 1 7, 2 1, 2 8, and 3 2, respectively). The sky brightness in each aperture is estimated from an annular region centered on each star with a width of 10 pixels (4 3) and inner radius of 14 pixels (6 0).

The output files from the IRAF scripts described above are used as input to our FORTRAN codes that perform ensemble differential photometry and proper error determination for each differential measurement on each star in the field. The procedure is an iterative one; an ensemble of bright calibration stars is chosen (hereafter called the “ensemble”), and a weighted mean magnitude is calculated for it in each exposure. We assume that temporal differences in this mean correspond to variations in atmospheric transparency and flux losses from the aperture because of changes in seeing. The changes in the ensemble’s mean magnitude are used to correct the instrumental magnitudes of each star to produce a precise differential light curve for each star in the field. Following the first iteration of this reduction, the light curves of the ensemble stars are examined statistically. Those ensemble stars found to be variable or having faults such as cosmic‐ray contamination in one or more exposures are removed from the ensemble and the process is repeated until all of the ensemble stars’ differential light curves are found to be nonvariable. Discussion of other high‐precision differential photometry techniques and results can be found in the literature (Howell, Mitchell, & Warnock 1988; Gilliland & Brown 1988, 1992; Newberry 1991; Honeycutt 1992; Gilliland et al. 1993; Merline & Howell 1995; Howell & Everett 2000; Everett, Howell, & Ousley 2000). Aside from the problem of ensemble differential photometry, a description of issues facing CCD photometry in general can be found in Stetson (1990, 1994).

Step by step, our reduction process is as follows:

1. First, we flag data points with an error if they represent stars lying too close to bad pixels or columns (closer than the aperture radius). Next, we apply a similar error flag to stars crowded by neighboring stars. After inspecting our fully reduced light curves, we were in position to determine how much crowding we will tolerate before assigning such an error. We conservatively assume that stars falling within 20 pixels of one another (8 6) are flagged with an error, unless their difference in magnitudes is large. If their magnitude difference is greater than 2, then only the fainter of the two stars is flagged with an error. Magnitude measurements flagged with an error (from proximity to bad pixels or crowding) are ignored in all of the data analysis described hereafter.

2. We group together those stars that occupy each pixel² half of our CCDs in order to reduce these stars with a “local” ensemble. This keeps the angular separation between the ensemble stars and stars of interest relatively small while including a sufficient number of bright, precisely measured stars that are candidates for the ensemble. The proximity of the ensemble stars to the stars of interest reduces the effects of differential atmospheric extinction across our wide field of view and noise from atmospheric scintillation. Scintillation noise reduction is discussed theoretically by Heasley et al. (1996), Ryan & Sandler (1998), and Dravins et al. (1997a), who suggested that by using local ensembles, it could be reduced by a factor of 3. In addition, Dravins et al. (1997b) note a color dependence to scintillation, showing that observations in red light are less adversely affected than blue and therefore limiting the bandpass is highly desirable. Our use of a nonblue, limited bandpass is an advantage; the V band provides the best compromise between the CCD quantum efficiency and redder bands for which variable night‐sky emission lines would add to the noise. Using a local ensemble also reduces differences in the stellar point‐spread functions within the group caused by optical aberrations, although by eye such problems were not seen in our data. Furthermore, spatial variations in obscuring clouds may also be treated more effectively under this scheme.

3. We first recalculate magnitudes and magnitude uncertainties for each star from the stellar and sky fluxes reported by IRAF (IRAF’s PHOT task calculates only rough errors). Here we adopt values for the gain and read noise of each CCD measured using our flats and bias frames. The uncertainty in a single magnitude measurement, , in magnitudes, is given by (see Howell 1993) where is the number of ADUs from the star in the aperture, g is the gain of the CCD in e⁻ ADU⁻¹, is the number of pixels in the aperture, is the number of pixels in the annulus around the aperture used to measure the sky flux, is the flux in ADUs per pixel from the sky, and R is the rms read noise of the CCD in electrons. We ignore the dark current of the MOSAIC CCDs, which is negligible, and the quantization in the number of photons detectable owing to the fact that the gain is ∼3 e⁻ ADU⁻¹ (Merline & Howell 1995).

4. A correction to the instrumental magnitudes of the stars in each exposure is found by calculating the mean instrumental magnitude, , of the N ensemble stars weighted by their individual uncertainties, . The residuals between this mean on each frame and the mean instrumental magnitude of the ensemble stars on all M frames is subtracted from the instrumental magnitudes, , of the individual stars to find the precisely corrected magnitude, m, of each star in the exposure (eq. [2]): (Note that the third term in eq. [2] is simply the same constant for all frames and serves only to reduce the absolute value of the correction.) The error in the correction, , is then calculated formally from the uncertainties in each individual measurement of the N ensemble stars:

Finally, is added to the for each magnitude measurement on the frame for the total uncertainty on each point in the light curve:

By selecting a properly bright set of stars for the ensemble, we ensure that the second term is significantly smaller than the first. Essentially, it is negligible.

5. Once the data points in the light curve have been corrected with differential photometry, we adjust all light curves by a single correction to place them on a standard magnitude scale. This is done by finding the V‐band magnitudes of the ensemble stars on a V‐band frame observed in succession with observations of the Landolt standard field at a similar air mass. The magnitude scale in V, as well as UBRI (see item 7), is accurate to ±0.1 mag. This accuracy is easily enough to distinguish different color stars and provides a good spectral type and rough indication of the luminosity class.

6. As mentioned above, we have little indication of whether a variable star has been included in the ensemble stars until we have completed one attempt at the differential photometry. To identify and remove any variable stars from the ensemble, we examine a few basic statistics calculated for each light curve. Those stars in the ensemble whose light curves exhibit unusually large standard deviations for their predicted errors and those with an unusually large value of χ² for the fit of a constant magnitude to the light curve are removed from the ensemble. Typically 10%–20% of the stars are removed, leaving 10–30 stars in the ensemble. Those rejected include real variables, crowded stars, stars near bad pixels, and stars with a cosmic ray falling in the aperture in one or more exposures. The entire differential photometry process is repeated iteratively, until a satisfactory ensemble selection is reached (e.g., when no variables remain in the ensemble). Typically, after a full reduction, the ensemble star light curves have standard deviations from their means of 2–4 mmag, depending on their brightnesses. The fact that the differential photometry correction is based on a large number of bright stars ensures us that when it is applied to the ensemble stars themselves, they will not have their natural noise or variations significantly suppressed by “self‐correction.”

7. The colors of stars in UBRI are found by matching stars in the different color images to those in the V‐band master image and applying the same IRAF scripts to measure fluxes and the other relevant information for each object. The Landolt standard star fields are observed at a similar air mass to the UBRI color images to place the magnitudes on the standard scale. Systematic uncertainties in the UBRI magnitude scales, like the V magnitudes, are probably ±0.1 mag.

The result of our data reduction is a file for each star and photometry aperture size listing the object’s light‐curve data: the V magnitudes, magnitude uncertainties, times of observation, and position on the CCD, among other error flags and documentation. Another file lists the UBVRI colors for each star.

4. RESULTS

After our data reduction, we have UBVRI colors and V‐band light curves for each star. Hereafter we will present and discuss results from one of our two fields. This survey area, our “field,” lies at , and contains the vast majority of the total number of stars we observed.

Before we analyze the light curves, we attempt to remove any data points whose value may be corrupted. The IRAF PHOT task reports on errors it encounters with centering the aperture, estimating the sky background, and measuring the stellar flux. Any single exposure of a star found to have an error reported by IRAF is discarded. Another problem that we eliminate is data points in error resulting from cosmic‐ray hits in the aperture. The counts from a cosmic ray add to that of the star, resulting in an anomalously high flux in one exposure. To reject data points corrupted by cosmic rays, we find the standard deviation from the mean magnitude for each light curve. If there are four or fewer data points in a light curve that are brighter than the mean magnitude by 3.5 times the standard deviation, we reject these points. The only exceptions are cases where the deviant data points are consecutive in time. This procedure eliminates most cosmic‐ray artifacts from nonvariable light curves while leaving those of variable objects unaffected. We would, however, reject any variations that occur in a single data point brightening an otherwise nonvarying source (a variation less than 10 minutes long).

In Figure 1 we plot the V‐magnitude distribution of all objects in our field as measured in our V‐band color exposure. The magnitude ranges from to , with a completeness limit at , where the numbers per magnitude flatten off. The magnitude limits are set by the saturation of the CCD on the bright end and the detection limit on the faint end. Since different exposures have different saturation and detection limits, in terms of magnitude, there are slightly different limits on the magnitude distribution of the stars for which we have complete versus partial light curves. The brightest stars with complete light curves are .

(22 KB)

Fig. 1.— Histogram showing the distribution of detected objects as a function of V magnitude. The survey spans the range –21 with the limits set by saturation and the faint detection limit. A total of 9150 objects is shown.

Open New Window

In Figure 2 we plot B−V versus V−I for each object detected in the BVI passbands. Shown on the plot as solid and dashed lines are the calibrated colors for main‐sequence and giant branch stars, respectively, taken from Cox (2000) and Leggett (1992). We also label the V−I colors expected for various main‐sequence spectral types along the top of the plot. Based on the colors, most of the stars in our survey are spectral types F5–M0, while some later and earlier types are seen. Additionally, the galaxy models of Bahcall & Soneira (1980) show that in our field and our magnitude range, only 5% of all stars are expected to be giants. A few objects lie significantly away from the stellar locus and may represent rare stellar classes (e.g., carbon stars or stars with strong emission lines), rapidly varying objects, or unresolved galaxies. We note that objects imaged with extremely red or extremely blue colors will not tend to appear on this plot as they are too faint to be detected in the B or I band, respectively.

(29 KB)

Fig. 2.— Plot of B−V vs. V−I for the objects detected in these three passbands. The colors expected for main‐sequence and giant stars are shown as solid and dashed lines, respectively. These colors are taken from Cox (2000) and Leggett (1992). Most of the stars in our survey field have spectral types between F5–M0, as can be seen by V−I colors of main‐sequence spectral types labeled along the top of the plot.

Open New Window

Each complete light curve consists of 200 data points spanning about 4.5 hr on each of the 5 nights. In total, we have 90% or more complete and error‐free V‐band light curves for 11,500 objects (many light curves have one or a few rejected data points because of cosmic‐ray hits). About 75% of these stars also have a complete set of UBVRI magnitude data, while the others are detected in only some of the filters. Each light‐curve data point has an associated uncertainty based on the calculations discussed in § 3. We can measure our photometric precision empirically as a function of magnitude by measuring the standard deviation of the data points from their mean magnitude in each light curve. For nonvariable stars, the standard deviation should match the calculated uncertainties. In Figure 3 we plot the logarithm of the standard deviation from the mean of the magnitudes in each light curve, ( ), versus the mean magnitude of the light curve as measured in our 2 8 radius apertures. Most of the points in Figure 3 represent light curves of nonvariable stars and cluster along a locus that gives our photometric precision as a function of magnitude. Variable sources will have higher and thus lie above this curve.

(31 KB)

Fig. 3.— Plot of the logarithm of the standard deviation measured in the light curves vs. the mean magnitude of the light curve (each data point represents one light curve). This yields an empirical measurement of our photometric precision per exposure as a function of magnitude. Since most stars are nonvariable, their standard deviations are a function of Poisson noise. The solid line represents the theoretical relationship between photometric precision and magnitude based on the Poisson statistics. The majority of our light curves, representing the nonvariable sources, are consistent with the theoretical relationship. The observed photometric precisions range from 0.0020 to 0.2 mag between and , respectively.

Open New Window

Our photometric precisions, as measured by the standard deviations of complete light curves, range from 0.0020 to ∼0.2 mag from to . The observed precision as a function of magnitude matches our expectations based on the proper error treatment discussed in § 3. This can be seen in Figure 3, where we plot the theoretical precision versus magnitude as a solid line alongside the data. To calculate this theoretical precision, we adopted the values of σ for each star and exposure as calculated according to equation (4). Then we determined the mean value of for each light curve. Finally, we took the square root of the mean of to represent the mean theoretical precision for a light curve. This weighting by σ² best matches the weighting of data points measured by the standard deviation.

Our precision versus magnitude is also a function of aperture size. A large aperture such as that used to obtain the data plotted in Figure 3 works best for the brightest stars. The main reason for this is that for bright stars, the dominant source of noise is the Poisson fluctuations in the number of detected photons from the star, even if the aperture is relatively large (see eq. [1]). For the faint stars, the dominant source of noise is fluctuations in the number of photons from the sky. By using a relatively small aperture to perform photometry on the faintest stars, we are able to reduce the sky flux, and its uncertainty, in the aperture. Therefore, the aperture size and its corresponding light curve we choose to analyze depends on the brightness of the star of interest.

One source of photometric error that we do not attempt to correct for is color dependence of extinction when comparing stars of widely different spectral type. In our case, the effective extinction coefficients are averages of the product of the instrumental response and stellar flux distribution across the (broad) V‐band filter. This effect and its magnitude have been discussed previously (e.g., Young et al. 1991; King 1952). When observations are taken over a range of air mass, these differences in the atmospheric extinction between the ensemble stars and stars of interest could produce artificial variations in the reduced light curves unless some attempt is made to correct for them using a “color term.” Additionally, variations in atmospheric transparency with time, independent of air mass, could have a similar effect. However, we do not see such variations in our light curves and, furthermore, estimate their magnitude to be only of order 1 mmag, which would make their detection difficult. For example, Young et al. (1991) give a formula for the differential extinction between two stars, (WRdC)M, where M is the air mass, W is a factor proportional to the bandpass width squared (0.03 for the V band), R is the difference in the extinction coefficients for the B and V bands, and dC is the magnitude difference in B−V between the ensemble and a star of interest. If the difference in B−V between our ensemble and star of interest were as large as 1 mag, and taking extinction coefficients of 0.2 for the B band and 0.13 for the V band, the differential extinction between star and ensemble would vary by 2 mmag over the course of a night’s observation as the air mass ranged from 1.0 to 1.9 (as it does in our light curves). In principle, this effect might be detectable in our light curves of the bluest or reddest stars, but in practice most of our stars and especially the brightest ones for which we have the most precise photometry fall in a narrow range of color (see Fig. 2). Also, only a few exposures are taken at the highest air masses, making the effect even more difficult to detect.

We note that for cases in which differential photometry needed to be performed over a wider range of spectral types or air mass, color term effects could be reduced by matching the colors of the ensemble to the stars of interest (e.g., using different colored ensembles to compare with similarly colored stars), observing at longer wavelengths (e.g., in R) where the atmospheric extinction curve is less steep with wavelength, or employing color term corrections after determining each star’s spectral energy distribution (Young et al. 1991).

Instead of examining the photometric precision per exposure, we can average together (bin) adjacent data points for higher precision per resultant data point with a loss of time resolution. If the individual data points are distributed with normal errors, and each data point in a given light curve has the same uncertainty, the precision per data point should improve by a factor of , where N is the number of adjacent data points being averaged. To demonstrate this, we take a 4.5 hr average of the data points in each light curve to produce new light curves that consist of one data point per night for 5 consecutive nights (4.5 hr is approximately the duration of an extrasolar planet transit; see § 5). We plot the standard deviation measured from these light curves versus their mean magnitude in Figure 4 (analogous to our plot in Fig. 3). The solid line shows our theoretical precision calculated in the same way as for Figure 3 and discussed above. The data points cluster around the theoretical precision, showing that we indeed can average together adjacent data points and achieve the expected improvement in photometric precision. The increase in precision is actually greater than the case of discussed above because each data point in the light curve has a different uncertainty and the average is dominated by those data with the smallest uncertainties (the standard deviation measured for the unbinned light curves and plotted in Fig. 3 treats each data point equally). There is also an increased scatter of the data at a given magnitude in Figure 4 as compared to Figure 3. This is due to larger random fluctuations in because there are only five data points per light curve here as compared to 200 data points for the unbinned light curves. The mean precisions in Figure 4 range from 0.00019 to 0.013 mag between and .

(40 KB)

Fig. 4.— Plot of the logarithm of the standard deviation measured in 4.5 hr time‐averaged light curves vs. the mean magnitude of the light curve (each data point represents one light curve). This yields an empirical measurement of our photometric precision as a function of magnitude (see also Fig. 3). The time‐averaged light curves measured here consist of 5 data points each, one per night, with each point representing the mean magnitude on that night. This can be compared to Fig. 3 where the precision per single exposure is displayed. The solid line represents the theoretical relationship between photometric precision and magnitude. The majority of our light curves, representing the nonvariable sources, are consistent with the theoretical relationship. The photometric precisions range from 0.00019 to 0.013 mag between and respectively. The binning of light curves into 4.5 hr segments is applicable to variations with these timescales such as the transits of extrasolar planets.

Open New Window

In Figure 5 we plot light curves for four objects found in our field. Each light curve is broken up into five different panels, showing each night of data. In each light curve there are two obvious gaps in the data. These gaps appear on the first night when UBRI color observations were taken in the middle of the time sequence and on the third night when several exposures were discarded because of telescope balance problems and poor tracking. Table 1 provides coordinates and color information for the objects whose light curves are shown in Figure 5.

(137 KB)

Fig. 5.— Four example light curves from the survey with individual nights shown in separate boxes from left to right. In panel a we plot the light curve of a bright, nonvariable source measured to have a standard deviation of 0.0020 mag. This represents our high‐precision limit for this data set. Panel b shows a light curve with 2% flux variations occurring over several days. Panel c shows higher amplitude periodic variations, presumably from an eclipsing binary star. Panel d shows a light curve with 5% flux variations on relatively short timescales. Note that the magnitude scales in the different panels of this plot differ. See the text and Table 1 for more discussion of these data.

Open New Window

TABLE 1 Selected Stars

Open New Window

In panel a of Figure 5 we show the light curve of a nonvariable and bright point source. This light curve is measured to have a standard deviation of 0.0020 mag. This result agrees exactly with the predictions based on our proper error calculation. The light curve in panel b is that of a low‐amplitude variable with most of the variation occurring from one night to another and with little variation within each night. This phenomenon is relatively common in our data (occurring in a few percent of the stars). This particular object has colors consistent with a mid‐K dwarf, which is a spectral type common in our field, and the variations are likely to be due to starspots (Henry, Fekel, & Hall 1995). The light curve in panel c shows high‐amplitude periodic variations and is almost certainly a short orbital period eclipsing binary star. The colors of this star are that of a early to mid‐G dwarf. The light curve in panel d is rather unusual: it shows variations of some 5% or less on timescales less than 1 hr. We could find no definitive periodicity for this light curve. The colors of this star are those of a mid‐F to early G dwarf. It will be easy to do follow‐up observations of any of these objects, including spectroscopy, to classify them (this is just one of many possible spin‐off studies that will be performed with our data set).

5. DETECTING EXTRASOLAR PLANETS

The large number of stellar light curves observed at high photometric precision with the time‐sampling rates in our data set have a number of applications. Among these is the search for transits of extrasolar planets. With our best per‐exposure photometric precision of 0.002 mag, it is possible to distinguish 2–3 hr transits of depth 0.003 mag from random noise. The detectability of a transit depends on the relative sizes of the planet and its parent star. At this threshold, a Jupiter‐sized planet is detectable transiting an A4 V or later type star, a Neptune‐sized planet detectable transiting an K7 V or later type star, and an Earth‐sized planet detectable transiting an M8 V or later type star. Of course, to detect planets, the low‐level intrinsic variations in stars must be understood well to distinguish transits from other phenomena. The data used to search for transits can be used to characterize stellar variability for this purpose. A survey for extrasolar planet transits is quite practical given a dedicated modest‐sized telescope and wide‐field CCD imager to observe a sufficient number of stars. A number of such surveys, both ground and space based, have been performed (Gilliland et al. 2000; Brown & Charbonneau 2000; Street et al. 2000; Borucki et al. 2001).

We note that fields with at least several times (∼5) the stellar density of those observed here can be used with our techniques. Giant planets and brown dwarfs in short‐period orbits around late‐type dwarfs will be preferentially detected, and the sizes and orbital periods of such objects can be found from their parent stars’ light curves. Given an intensive observing program, wide‐field surveys for transits will provide a good statistical sample to determine the size distribution, orbital periods, and types of stars that harbor extrasolar planets. A transit search is also capable of detecting terrestrial‐sized planets with masses too small to be detected by other methods (such as the detection of radial velocity perturbations in stars).

6. CONCLUSION

We have presented results from a wide‐field photometric survey. The data consist of UBVRI colors and V‐band light curves sampled every 5.7 minutes for stars in a 1 deg² field. The colors of these stars show that most are dwarfs with temperature classes between F5 and M0, as expected. Our ensemble differential photometry of these stars yields light curves with photometric precisions as high as 0.0020 mag per exposure or 0.00019 mag per night, if we produce 4.5 hr averages.

Typical “transit times” are a few hours for extrasolar planets with short orbital periods. With these photometric precisions it is possible to detect transits by Jupiter‐sized planets in orbit around A4 V or later type stars, transits by Neptune‐sized planets in orbit around K7 V or later type stars, and transits by Earth‐sized planets in orbit around the M8 V or later type stars. An analysis of the stellar variability and search for transits in the present data will appear in a follow‐up publication.

Using the techniques outlined in the paper, we are beginning a photometric survey program with the 1.3 m Robotically Controlled Telescope located on Kitt Peak. Our CCD camera will be better suited for high‐precision photometric work, yielding photometric precisions that are 5 times better than those discussed here. A major component of the program is the search for planetary transits.