SPECKLE INTERFEROMETRY AT THE BLANCO AND SOAR TELESCOPES IN 2008 AND 2009

The observations reported in this paper were obtained with the high-resolution camera (HRCam)—a fast imager designed to work at the SOAR telescope, either with the SOAR Adaptive Module or as a stand-alone instrument. The HRCam is described by Tokovinin & Cantarutti (2008); here we recall its main features.

HRCam uses a CCD detector with internal electron multiplication—an EMCCD. The Luca camera from Andor⁴ was chosen for its low cost, fast frame rate, and simple signal interface via a USB port. The CCD has 658 × 496 10 μm pixels. It is cooled thermoelectrically to −20°C, resulting in a very low dark current for the short exposure time used here (except in a small number of hot pixels). We used an electron multiplication (EM) gain of 44, so the readout noise of 14 electrons is effectively reduced to 0.3 el. The quantum efficiency of this detector is around 0.5.

HRCam consists of the detector, mechanical structure, filter wheel, and optics. The f/16 beam coming from SOAR is collimated by a 50 mm negative achromat (Barlow lens) and refocused by a 100 mm positive lens, doubling the effective telescope focal length and providing a pixel scale of 15 mas. At the Blanco f/8 telescope, we replaced the negative lens by a positive achromat with 25 mm focal length to get adequate sampling.

HRCam has no corrector for atmospheric dispersion (AD). Most observations were obtained through a Strömgren y filter, while brighter stars were also observed through an Hα interference filter, especially at large zenith distances. A few observations were taken through standard wide-band V, R, or I filters. Table 1 lists the central wavelength and bandpass of each filter as measured (except for the I filter where the bandpass is determined by the detector cutoff in the red).⁵

The camera software developed by R. Cantarutti works on a PC computer under the Linux operating system. All basic functionality is provided, including the filter wheel and detector control, and connection to the telescope control system. We used the detector in free-run continuous mode. The central region of 200 × 200 pixels may be read out with an exposure time of 20 ms at a rate of 31.8 ms per frame, or a larger 400 × 400 region may be read at a 43.1 ms frame rate. The required number (typically 400) of frames is grabbed during 13–17 s and written to the disk. Immediately after the acquisition, a quick-look power spectrum (PS) is calculated and displayed, showing the quality of the data and the fringes for resolved binaries.

Observing runs are listed in Table 2. As noted above, the use of HRCam at the Blanco telescope was not foreseen. The software was not interfaced to the telescope, so information missing in the FITS headers (acquisition time, object coordinates, zenith distance) was retrieved later using our logbook. A few object identification errors were corrected post factum, but some may still remain undetected. We planned to use the ADC of the Blanco telescope, but found it to be nonoperational, so the dispersion remained uncorrected and the program was restricted to smaller zenith distances. All five nights of the Blanco run were clear, with variable seeing.

The SOAR telescope is located at the Cerro Pachón mountain. The shape of its thin 4.1 m mirror is controlled actively using a bright star at the beginning of the night and using look-up tables for the rest of the night. HRCam was installed at the Nasmyth focus and received the light after three reflections. SOAR has an alt-azimuth mount; field rotation at the Nasmyth focus is compensated by counter-rotating the instrument. The position angle of the rotator is recorded in the headers for subsequent calculation of angles on the sky.

In preparation for the SOAR runs, we tested the system during a technical night on 2008 August 8. Some useful data were collected during this night, mostly on binaries with known orbits. During the two scheduled three-night runs in 2008 October and 2009 April, the sky was clear, the wind speed was low, and the seeing was generally good, with the width of the best long-exposure images as small as 070.

Efficient use of the allocated telescope time required good preparation of the observing program and a substantial effort on the part of the telescope operators. For example, during the April run, we observed 551 objects on three nights, spending less than 3.5 minutes per star on the average. Each observation included telescope slew, object acquisition, data recording, and quick-look analysis. We recorded at least two data cubes of each object and analyzed each cube independently. This strategy helped to verify new companion detections and to estimate observational errors. The last column of Table 2 lists the total number N_dat of data cubes taken during each run.

The faintest stars observed were around V = 10^m, with a few exceptions. For example, the components of DON 93 BC, V = 12^m and 13.1^m at 076 separation, were measured reliably with an accuracy of 1 mas under good seeing. Companions of V ∼ 13^m were measured routinely if the primary star was brighter than 9^m. Paradoxically, the presence of a bright primary increases the sensitivity for the secondary because the speckle signal is proportional to the product of photon fluxes from both companions. For faint stars, sensitivity could be gained at the expense of resolution by increasing the exposure time (up to 100 ms) and by observing in wide-band filters. This last resource was, unfortunately, not available to us because it requires a working compensator for AD.

Each data cube typically contains K = 400 images of 200 × 200 or 400 × 400 format, with 14 bits per pixel. The average dark frame (with the same exposure time), called bias, is subtracted from each image to remove the fixed offset of ∼505 analog-to-digital units (ADU) and the dark current. The average bias is stable to within 1 ADU. Pixels in the bias frame which are more than 10 ADU above the average level are identified as "hot;" these pixels in the images are replaced by the average of neighboring pixels. Finally, all pixels below 17 ADU (twice the readout noise) are set to zero in order to reduce the influence of the remaining pattern and readout noise. The optimum threshold value was selected after some trials and then applied to all data. Single photons produce signals well above this threshold. Despite the thresholding and the very low dark current, each 200 × 200 frame contains some 40 photon events at random locations, apparently created in the electron-multiplication process. This additional background is the main factor which prevents observing very faint stars with long accumulation times. For a subset of bright stars observed without EM gain, only a fixed bias of 505 ADU was subtracted, without any thresholding.

The PS of an image cube is calculated by summing the square modulus of the Fourier transform of each image:

$$\begin{equation} P(f_x,f_y) = \frac{C}{K}\; \sum _{i=1}^{K}| \tilde{I}_i(f_x,f_y) |^2. \end{equation} \tag{ 1 }$$

The spatial frequencies f_x, f_y correspond to the elements of square discrete arrays. The normalization constant, C, is determined from the condition P(0, 0) = 1. In the following, it is convenient to use normalized frequencies κ = f/f_c, where f_c = D/λ is the cutoff frequency, D is the telescope diameter, and λ is the central wavelength of the filter passband.

It is well known that the PS of a single bright star P₀(f_x, f_y) contains two components: a strong signal at low spatial frequencies f < r₀/λ which corresponds to the seeing-limited image (r₀ is the Fried parameter) and a high-frequency component extending up to f_c and produced by the speckle structure. An additive component P_noise is produced by the photon noise. It is easily estimated by averaging the PS values at f>f_c. This additive term is then subtracted from the PS.

Knowledge of the speckle transfer function (STF) P₀ is needed for fitting a binary-star model to the data (see below). Single reference stars are sometimes observed for this purpose. However, the STF is not stable in time and depends on such factors as seeing conditions, AD, telescope aberrations, vibrations, etc. For each PS, we describe the STF in the high-frequency zone by an empirical model, obviating the need for a reference star and accounting for the changing conditions automatically. The bias term P_noise is subtracted from P(f_x, f_y), then the PS is averaged azimuthally, leading to the one-dimensional function $\overline{P}(\kappa)$. A very simple two-parameter model,

$$\begin{equation} \log _{10} P_r(\kappa) \approx \log _{10} [\overline{T}_{{\rm AD}}(\kappa) T_0(\kappa)] + p_0 + p_1 \kappa, \end{equation} \tag{ 2 }$$

is fitted in the range κ_min < κ < κ_max. Typically, we select κ_min = 0.2 and κ_max = 0.8, but for the noisy data the upper limit is reduced. Here $T_0(\kappa) = 2/\pi [ \arccos \kappa - \kappa \sqrt{ 1 - \kappa ^2 } ]$ is the diffraction-limited transfer function of an ideal telescope (the central obstruction is ignored), and $\overline{T}_{{\rm AD}}(\kappa)$ is the azimuthally averaged deterministic blur T_AD(f_x, f_y) caused by the AD,

$$\begin{equation} T_{{\rm AD}}({\bf f}) \approx \exp [ - 2 \pi ^2 ({\bf f} {\bf x} /2.506)^2 ]. \end{equation} \tag{ 3 }$$

The AD blur is represented by a Gaussian function. The length of the blur vector |x| = [n(λ₁) − n(λ₂)]/p tan z (in pixels) is known for the zenith distance z, refractive index of air n(λ), filter bandwidth limits λ₁ and λ₂, and the pixel size p. The direction of the vector x is known from the calculated parallactic angle and the detector orientation. Figure 2 illustrates PS modeling.

Zoom In Zoom Out Reset image size

The parameter p₀ shows the level of the high-frequency component of the PS extrapolated to zero frequency. The theoretical PS model predicts that $(D/r_0)^{-2} = 0.435\, 10^{p_0}$, leading to an estimate of the seeing conditions relevant to each data cube from the p₀ values. These seeing estimates match quite well the half-width of the re-centered long-exposure images calculated from the data cubes. The second parameter, p₁, shows how fast the high-frequency component of the PS is decreasing. For an ideal speckle pattern p₁ = 0, but in reality finite exposure time, finite bandwidth and other factors lead to p₁ < 0.

The synthetic STF is thus calculated as

$$\begin{equation} P_{0, syn}({\bf f}) = T_{{\rm AD}}({\bf f}) T_0(f) 10^{p_0 + p_1 (f/f_c)} \end{equation} \tag{ 4 }$$

for the selected frequency range and f = |f|. Alternatively, we use the azimuthally averaged observed PS $\overline{P}(f)$ as a reference, with an additional multiplier $T_{{\rm AD}}({\bf f})/ \overline{T}_{{\rm AD}}(f)$ to account for the AD. For binaries with separations above 01 the radially averaged reference is normally chosen, while synthetic reference is used for closer binaries.

The PS of a binary star shows characteristic fringes. It is more practical, nevertheless, to detect companions in the auto-correlation functions (ACFs) calculated from the PS by Fourier transform. The two-component structure of the PS is carried to the ACF which consists of a wide seeing pedestal and three narrow peaks (in the case of a binary star). The pedestal can be removed by setting to zero the PS at low spatial frequencies, e.g., at f < 0.2f_c. Such crude filtering leads to "ringing" in the ACF. To avoid it, we divide the PS by its azimuthal average $\overline{P}(f)$ at low frequencies where it exceeds the extrapolated level of the speckle signal, $10^{p_0}$, and apply additional Gaussian damping to further reduce the low frequencies. The filtered ACFs are then computed from the filtered PS and are used together with the PSs for binary-star analysis.

The parameters of a binary star are the time of observation, T, separation, ρ, position angle, θ, and magnitude difference, Δm. The first number is arbitrarily precise, the next two numbers are combined in a two-dimensional vector r = (ρcos θ, ρsin θ). The observed PS P(f) (after subtraction of P_noise) is fitted by a model

$$\begin{equation} P_{\rm mod}({\bf f}) = P_0({\bf f})[ A + B \cos (2 \pi {\bf f} {\bf r}) ], \end{equation} \tag{ 5 }$$

where P₀(f) is the STF and the coefficients A and B are related to the magnitude difference. The position angle is determined only modulo 180°, so a change of quadrant is always possible.

Fitting of model (5) to the PS is done by the Levenberg–Marquardt method in the frequency range κ_min < κ < κ_max over the upper half-plane κ_y ⩾ 0, due to the symmetry. The error of the measured PS at each point j, σ²_j = (P_j + P_noise)/K is taken into account (K: number of images in the data cube). The quality of the fit is evaluated by the normalized sum of residuals χ²/N = (1/N) ∑_j(P_j − P_mod,j)²/σ²_j, N being the total number of the fitted points in the frequency plane. For noisy data, we obtain χ²/N ∼ 1, but for bright stars the un-modeled systematic features of the STF (e.g., details caused by telescope aberrations) dominate the residuals, leading to χ²/N values up to 20. Figure 3 shows the example of a binary-star PS and its fitted model. The fact that AD is explicitly included in the model helps to distinguish true close companions from the elongation of speckles produced by the AD.

Zoom In Zoom Out Reset image size

A model of a resolved triple star is fitted in a similar way, but the number of parameters is larger. Initial values of the fitted parameters are determined by clicking on the companion(s) on the displayed ACF. In the data table, we list positions and magnitude differences of triple-star pairs relative to the brightest component, not the photo-centers of close sub-systems. For example, the measurements of J01198 − 0031 = STF 113 A,BC at 167 in fact refer to the pairing (A,B), not to the center of the inner sub-system FIN 337 BC. Accordingly, in the data tables we designate the wide pair as STF 113 AB. The quadrants in a triple system can be changed jointly, but not individually. When the quadrant of the slowly moving outer pair is known from previous observations, the quadrant of the inner pair can be established without ambiguity.

The fitting program provides estimates of the parameter errors. Yet another estimate comes from the variance of N measurements obtained on the same night in the same filter, $\sigma _x^2 = (N-1)^{-1} \sum _N (x - \overline{x})^2$. Mostly, N = 2 (two data cubes). We adopt the larger of these two errors and list them in the data table. For 1846 measurements with N ⩾ 2, the median errors of companion positions are 0.3 mas in both radial and tangential directions, while 75% of the errors are smaller than 0.8 mas. For a subset of 49 measurements with N ⩾ 4 where the estimates of the variance are more reliable, the median errors in tangential and radial directions are 0.5 mas and 0.6 mas, respectively. The error in separation exceeds 7.5 mas (half-pixel) in only 16 cases where the companions are either close or very faint. The listed errors are internal; they do not take into account calibration uncertainties or other systematic effects. During the Blanco run, some bright pairs were observed repeatedly, and the agreement between these measurements is quite good. For example, A 417 (WDS 23052 − 0742) shows the scatter σ_ρ = 0.30 mas and ρσ_θ = 0.23 mas from six measurements over four nights in two filters.

Accurate knowledge of the detector pixel scale and orientation is needed to convert binary-star parameters from fitted values in pixels to absolute positions on the sky. Speckle measurements at the Mayall 4 m telescope at Kitt Peak National Observatory were calibrated by means of a double-slit mask (Hartkopf et al. 2000), while speckle data at the Blanco telescope were traditionally tied to this calibration by observing common pairs. Originally, we intended to observe many binaries with known orbits to calibrate our runs. However, it turned out that the quality of the available binary-star orbits is not adequate. By calibrating against orbits, we relate modern precise measurements to the historical data of much lower accuracy which may also contain systematic errors.

A comparison between the Blanco08 and SOAR08b runs has revealed a disagreement of the pixel scale at the 3.5% level, despite independent calibration of each run with ∼100 orbits. In the face of this discrepancy, we calibrated the SOAR09 run by projecting into the telescope a fringe pattern formed by two coherent point sources attached to the telescope spider. The baseline of this interferometer b = 0.4999 m was accurately measured, the wavelength of the green laser λ = 532.2 nm is known, so the fringe period λ/b=02196 is known as well. About 30 fringes fit into the 400 × 400 pixel field. The position of the fringe peak in the PS of these data cubes is found by a simple centroid, leading to the determination of the pixel scale and detector orientation.

The results for each of the eight series of fringe measurements are very consistent internally, but do show a spread between the series amounting to 0.5% in scale and 02 in angle. We attribute these differences to small imperfections in the fringe pattern caused by optical defects and aberrations in the beam path. The pixel scale of 15.23 mas is finally adopted for all SOAR runs. We measured also the effective pixel size of the HRCam detector (through its optics) by illuminating the device with a laser and determining the angle between the beams diffracted back by the pixel grid. The nominal projected pixel size of 5.00 ± 0.025 μm was confirmed. Using the effective focal length of the SOAR known from its optical prescription, F = 67.834 m, we obtain the pixel scale of 15.20 mas in agreement with the laser calibration.

The detector orientation in the SOAR09 run was independently checked by observing stars at large zenith distances with wide filters. The resulting PS is elongated perpendicularly to the AD direction which itself is known. The elongation angle is determined by correlating the observed and modeled PSs in a certain range of spatial frequencies and finding the angle where the correlation reaches maximum. It turned out that most consistent results are obtained by considering the mid-range spatial frequencies between 0.1 and 0.3f_c, while at higher frequencies the direction of the elongation is possibly affected by telescope vibrations (see below). The resulting detector angle determined from AD is 132 ± 01, to be compared to 137 ± 01 measured from fringes and 152 ± 02 from orbits. In this case, all three methods agree very well.

The pixel scale of the Blanco run was adjusted using 107 binaries measured also in 2008 October (Figure 4). The unweighted rms difference ρ_Blanco − ρ_SOAR after adjustment is only 3.5 mas. A similar level of agreement is seen between other pairs of runs. This is the upper limit for the external measurement errors (part of the difference is caused by the motion of binaries between the runs). In contrast, the residuals in ρ to the orbits (Figure 5) show a large scatter for the whole data set and for the individual runs, precluding accurate scale calibration. Calibration only with the orbits of grade 1 does not help. The rms scatter of the O − C residuals in ρ in Figure 5(b) is still 8.3 mas for 32 points with ρ < 04, much larger than the difference between the runs. The residuals to orbits in the tangential direction are around 5 mas.

Zoom In Zoom Out Reset image size

Zoom In Zoom Out Reset image size

The detector orientation changes slightly at each installation of the camera, so it has to be calibrated for each run. Considering the comparisons with orbits, the difference in θ between wide binaries measured commonly in pairs of runs, and the secure angular offset determined for the SOAR09 run, we adjusted the offsets in θ iteratively to reach mutual consistency between all runs. The remaining average differences in θ between any pair of runs are less than 02, while the pixel scales are consistent to within 0.2% or better. We believe that the absolute calibration errors do not exceed 1% in scale and 05 in angle, and that they are likely smaller. The data presented in this paper are possibly the most accurate measurements of southern binaries done so far.

The contrast of fringes in the PS β = B/A (or the ratio of peaks in the ACF β/2) is related to the magnitude difference between binary components Δm,

$$\begin{equation} \Delta m = -2.5 \log _{10} [(1 - \sqrt{ 1 - \beta ^2})/ \beta ]. \end{equation} \tag{ 6 }$$

For small Δm, the slope of this relation is shallow, leading to a larger error of relative photometry. Moreover, as the fringe contrast is often underestimated, the Δm is overestimated. The positive bias on Δm becomes significant for faint stars, where the PS models fail. It is likely that the background photons cause this effect, given that PS is related to the intensity in a nonlinear way (Equation (1)). For faint stars, the slope of the PS models |p₁| is systematically less than normal, indicating that something is wrong.

We compared our speckle photometry with that of Hipparcos (ESA 1997), using only the Strömgren y data as that filter most closely matched the Hipparcos Hp filter. We found that the positive bias on Δy is strongly correlated with the signal-to-noise ratio (S/N) δ, which we define here as the ratio of the speckle signal to the photon-noise bias at half of the cutoff frequency:

$$\begin{equation} \delta = T_0(0.5) \; 10^{p_0 + 0.5 p_1} {/} P_{\rm noise}. \end{equation} \tag{ 7 }$$

Figure 6 shows such correlation for the Blanco run. Observations with δ < 0.25 are marked by colons in the data table. These biased estimates are still useful as upper limits, especially for close pairs where no other photometry is available.

Zoom In Zoom Out Reset image size

Another reason for Δm over-estimation is the loss of correlation between speckle patterns of wide binary components (anisoplanatism). We implemented an alternative scheme for estimating Δm directly from average re-centered images, provided that the binary is resolved (ρ larger than the half-width of long-exposure image). The relative position of the components is already known from the PS fitting, so we have to determine only the true quadrant and Δm. The quadrant is selected by comparing two point-spread functions (PSFs) obtained by de-convolving the average image from the binary. For the wrongly selected quadrant, the PSF has a negative lobe opposite to the companion, so we select the quadrant with the least negative PSF. The de-convolved PSFs are then computed for a grid of Δm values from 0 to the Δm estimated from speckle. The final Δm value is the one which gives the most symmetrical PSF, when the secondary component disappears. This procedure is applied automatically to all data, but in some cases (wrong quadrant choice for Δm ∼ 0 or large Δm) it fails. In the following, we call this method resolved photometry and mark such Δm estimates to distinguish them from the standard speckle photometry of closer pairs.

Figure 7 compares Δy measured by speckle (only data with δ>0.25) and by resolved photometry with the Hipparcos photometry, for the whole data set. These plots help to evaluate the accuracy of our photometry. Despite some remaining deviant points, the overall agreement is evident. Wide pairs without resolved photometry suffer from the positive Δy bias caused by anisoplanatism. As this bias is variable, depending on high-altitude turbulence, we do not attempt to quantify it.

Zoom In Zoom Out Reset image size

Quantitative evaluation of the bias and precision of our photometry is given in Table 3. We compare speckle photometry of close (ρ < 1'') pairs with good S/N (δ>0.25) and resolved photometry of wide pairs with magnitude differences ΔHp measured by Hipparcos (first two lines) and with magnitude differences ΔV derived from the Tycho data by Fabricius & Makarov (2000) (last two lines). Each line lists the number N of pairs in common, median, and average difference between Δm's, and the rms dispersion of the difference. The speckle photometry has a small positive bias, while the resolved photometry is essentially unbiased. The speckle photometry bias is larger in comparison with the Tycho ΔV than with ΔHp because only pairs wider than 03 are considered by Fabricius & Makarov (2000), hence larger contribution of anisoplanatism. When we compare our Δy with ΔHp only for pairs with 03 < ρ < 1'', a similar positive bias of 0.2^m appears, but the rms scatter becomes smaller, around 0.20^m. The Hipparcos photometry of close pairs with ρ < 02 (below the resolution limit of the Hipparcos telescope) is suspect, as can be seen in the lower plot of Figure 7. We estimate that intrinsic random errors of our photometry (both speckle and resolved) are around 0.2^m rms, but in a few cases the disagreement between our and published photometry is much larger.

Binary companions are detected as symmetric spikes in the filtered ACF. Fluctuations in the ACF caused by photon noise, residual speckle statistics, etc. prevent detection of faint companions. The rms fluctuations in the annuli of 2 pixel width around the central peak are calculated for each ACF; this σ(ρ) curve is translated to the detection limit Δm₃(ρ) by assuming that all companions above 3σ are detectable. Of course, the annulus containing the actual companion will have enhanced fluctuations and lower Δm₃. However, as shown in Figure 8, a simple linear model can be fitted to the Δm₃(ρ) curve by excluding the companion zone. Typically, the slope of the curve changes abruptly at some distance ρ* ∼ 015, so we approximate it by two linear segments intersecting at ρ*. Such two-segment linear models are fitted to all data.

Zoom In Zoom Out Reset image size

The detection threshold was checked by simulating fake companions. A real ACF of a single star (or a binary de-convolved from a faint companion) was used as a model of the speckle PSF; then companions were generated with separations from 01 to 1'' and Δm in the (−1.0, + 0.5) interval around Δm₃. About 10 representative cases were tested in this way, with 100 trial companions each (Figure 9). Our general conclusion is that the 5σ line Δm₅ = Δm₃ − 0.55 corresponds to certain detection, companions in the region between Δm₅ and Δm₃ are detected fairly frequently, and companions with Δm>Δm₃ remain undetected, with few exceptions.

Zoom In Zoom Out Reset image size

Figure 10 compares the detection limits estimated by the above procedure with the actually measured Δm. Only data with good signal-to-noise ratio δ>0.25 and ρ>015 are selected. Positive bias in Δm inherent to speckle photometry is also relevant to the detection limits which are over-estimated by the same amount. For wide companions with ρ>1'' anisoplanatism becomes important, making our formal detection limits optimistic. The same is true for the noisy data with δ < 0.25. We list the Δm₅ detection limits for unresolved targets at separations of 015 and 1'' and mark cases with δ < 0.25 by colons. The actual detection limits for companions closer than 01 cannot be established by the above simple analysis, as they depend on a number of artifacts discussed in the following sub-section. Median detection limits Δm₅ for the whole data set are 4.22^m and 5.33^m at 015 and 1'', respectively. For the best 25% of data, these limits exceed 4.67^m and 6.08^m.

Zoom In Zoom Out Reset image size

In all data sets, most ACFs are round, but some ACFs show symmetric enhancements near the first diffraction ring which can be mistaken for a binary companion with Δm ∼ 3. In the case of binary stars, these false details appear around the secondary peaks as well, distinguishing them from true triple systems where the secondary peaks are doubled, not tripled.

These false peaks have some common features. First, their separation from the center, typically from 45 mas to 75 mas, is larger in the Hα filter than in the y filter, while the intensity of the peaks is also larger in Hα. Second, the peaks are almost always oriented vertically, along the AD direction. Yet, they are not caused by the AD because the separation does not depend on the spectral bandwidth and the zenith distance (the peaks are seen even near the zenith). Third, the separation, orientation, and intensity of the peaks are variable. In two data cubes of the same star, one may have the false peaks while the other does not. When we split the data cube into 10 segments and calculate the PS for each segment, the variability of the peaks on a timescale of seconds becomes even more apparent. However, the peaks often appear persistently in the ACFs of different objects observed one after another in the same part of the sky.

The persistent nature of these peaks means that they are not caused by random fluctuations of speckles and do not disappear when more data are averaged. Having at least two data cubes for each object and examining data on other objects observed before and after the star usually helps to identify and reject false companions, despite their striking resemblance to real binaries in some cases.

The properties of the false peaks indicate that they are likely caused by variable optical aberrations with a characteristic size of 2 m, or $\frac{1}{2}$ of the telescope diameter. We simulated speckle data by adding a sinusoidal wave-front aberration with $\frac{D}{2}$ period to the atmospheric distortions. Some characteristics of the false peaks (Figure 11, right) could be reproduced. The intensity of the false peaks is larger in Hα than in y, and it decreases with degrading seeing. Orientation of the peaks in the vertical direction suggests that optical aberrations such as astigmatism could play some role, but our simulations show that the STF can be affected only by a fairly large amount of defocus and astigmatism causing visible elongation of the seeing-limited PSF. Even then the astigmatism produces an elongation of speckles, rather than their tripling. Air stratification in the dome can possibly cause this optical effect, but its exact nature remains mysterious. Such false peaks could explain previous detections of speckle companions which turned out to be bogus. See the discussion of false speckle companions by McAlister et al. (1993).

Zoom In Zoom Out Reset image size

Sometimes speckle peaks in the ACF are also elongated at a large angle with respect to the AD. This blur could be caused by telescope aberrations or tracking errors. Although tracking errors are usually slow (typical frequency 1 Hz), their amplitude can be large enough to degrade the resolution in a 20-ms exposure. Alternatively, speckles can be elongated by fast turbulence. Instrumental elongation of speckles is indistinguishable from the effect of a close binary companion at the limit of resolution (∼30 mas), so only the examination of data on stars observed before or after can help to distinguish an authentic close binary from an artifact.

Clearly, detection and measurement of close binary companions are complicated by the artifacts and involve an element of human judgment and error. The detection limits cannot be formalized, as was done for wider companions. We cannot exclude the possibility that some of the measurements presented below are affected by the artifacts, despite all efforts to understand and eliminate them.

Table 4 lists 1898 measurements of 1189 resolved known and new binary stars and sub-systems. Its columns contain (1) the Washington Double Star Catalog (WDS; Mason et al. 2001b) designation, (2) the "discoverer designation" as adopted in the WDS, (3) an alternative name, mostly from the Hipparcos catalog, (4) Besselian epoch of observation, (5) filter, (6) number of individual data cubes, (7,8) position angle θ in degrees and internal measurement error in tangential direction ρσ_θ in mas, (9,10) separation ρ and its internal error σ_ρ in mas, and (11) magnitude difference Δm. An asterisk follows the value if Δm and the true quadrant are determined from the resolved photometry; a colon indicates that the data are noisy and Δm is likely over-estimated. We decided not to mark with colons the Δm values of wide pairs over-estimated only due to anisoplanatism (when no resolved photometry is available), to avoid confusion with the low S/N cases. Note that in the cases of multiple stars, the positions and photometry refer to the pairings between individual stars, not with photo-centers of sub-systems.

For stars with known orbital elements, Columns 12–14 of Table 4 list the residuals to the ephemeris position and the reference to the orbit from the Sixth Orbit Catalog (Hartkopf et al. 2001). In those cases where multiple orbits for the same system are present in the catalog, the orbit with the smallest residuals is selected. An asterisk in the final column indicates that a note concerning this system may be found in Table 7.

Table 5 contains the data on 285 unresolved stars, some of which are listed as binaries in the WDS or resolved here in other runs. Columns 1–6 are the same as in Table 4 (although Column 2 also includes Bayer designations, HD numbers, or other names for objects without discoverer designations). Columns 7 and 8 give the 5σ detection limits Δm₅ at 015 and 1'' separations determined by the procedure described above. When two or more data cubes are processed, the best detection limits are listed. Noisy data with δ < 0.25 are marked by colons to indicates that the actual detection limits are smaller. As in Table 4, the final column indicates a note to the system.

Table 5. Unresolved Stars

WDS (2000)	Discoverer	Hipparcos	Epoch	Filter	N	5σ Detection Limit		Δm	Note
	Designation	or Other	+2000			Δm(015)	Δm(1'')	Flag
	or Other Name	Name				(mag)	(mag)
00024 + 1100	HD 224983	HIP 184	8.5378	y	2	4.06	4.63
00059 + 1814	LTT 10019	HIP 493	8.5377	y	2	4.03	4.49
00063 − 4905	HDO 180	HIP 522	8.5379	y	2	4.60	6.19		*
00084 + 0637	HD 377	HIP 682	8.5378	y	2	4.60	5.46
00113 − 1528	6 Cet	HIP 910	8.5379	y	2	4.96	6.19
00116 + 1020	HD 727	HIP 943	8.5378	y	2	3.99	4.55
00117 − 3508	the Scl	HIP 950	8.5379	y	2	4.76	6.27
00125 + 1434	LN Peg	HIP 999	8.5378	y	2	3.66	4.19
00174 + 0853	STF 22 C	HIP 1392	8.7672	y	2	4.06	5.84		*
00201 − 6452	zet Tuc	HIP 1599	8.5379	y	2	4.60	6.01

Only a portion of this table is shown here to demonstrate its form and content. Machine-readable and Virtual Observatory (VO) versions of the full table are available.

Download table as:  Machine-readable (MRT)Virtual Observatory (VOT)Typeset image

New discoveries are repeated in Table 6 in the same format as measurements in Table 4—a total of 48 pairs. Figure 12 shows ACFs of 20 newly resolved triple systems.

Zoom In Zoom Out Reset image size

Notes to some objects in Tables 4–6 are given in Table 7. These notes include miscellaneous information such as additional components, discovery history, etc. The WDS (Mason et al. 2001b) and the Multiple-Star Catalog (MSC; Tokovinin 1997) were extensively consulted, among other sources. Each system is identified by its WDS designation and an alternate name. Cases where deviations from the orbits are quite large are also indicated. The definition of unacceptably large residuals is subjective; we consider as such orbits which deviate from our measurements by more that 20° in θ or by more than 50% in ρ. There are 131 such cases out of 544 systems with orbits. The 24% fraction of bad orbits demonstrates that more measurements of southern binaries are needed.

Table 7. Notes to Individual Systems

WDS (2000)	Discoverer	Note
	Designation
	or Other Name
00028 + 0208	BU 281 AB	C at 44'' is optical
00059 + 1805	STF 3060 AB	CPM in WDS
00059 − 3020	RST 5180 AB	dy = 1.2, dm (WDS) = 0.2. The status of C at 5'' is unknown
00063 − 4905	HDO 180	Companion at 4'', outside field. A has a planetary companion
00098 − 3347	SEE 3	Bad orbit
00121 − 5832	RST 4739	dm (Blanco) too large, WDS: dm = 0.17 mag
00174 + 0853	A 1803 AB	WDS lists three more companions, but only C at 34 is physical (MSC)
		Two orbits for AB. C component itself is unresolved
00271 − 0753	A 431	dy = 0.6, dm (WDS) = 0.11
00310 − 1005	BU 1158 BC	The orbit by Baz1991a does not match, predicting 013
		separation. Component A at 79'' is optical. On 2008.5460,
		we obtained a very different measure (3562,  0.0475, '' Δy = 0.8),
		presumably pointing a wrong object (maybe the component A?)
00315 − 6257	I 260 CD	AB at 27'' from CD is physical (MSC)
		in this system beta 1,2 Tuc containing six known components

Only a portion of this table is shown here to demonstrate its form and content. Machine-readable and Virtual Observatory (VO) versions of the full table are available.

Download table as:  Machine-readable (MRT)Virtual Observatory (VOT)Typeset image

In this sub-section we give more lengthy comments on a few selected cases.

02053-2425 = HIP 9774 = I 454. The brightest companion of I 454 (also known as ADS 1652) is a double-lined spectroscopic binary with period 2.6 years and eccentric orbit, e = 0.78 (A. Tokovinin 2010, in preparation). The estimated semimajor axis of this pair is 47 mas. The system passed through periastron in 2008 May–June and was marginally resolved in 2008 July at Blanco. In 2008 October the separation was closer, below the diffraction limit of the 4 m telescope. Nevertheless, we were able to fit consistently a triple-star model to seven power spectra recorded in October. As our estimated Δm ∼ 1^m is larger than the spectroscopically estimated Δm ∼ 0.3^m, it is possible that the actual separations were even smaller than those listed in Table 6. Component C = HIP 9769 of this multiple system was also observed here and found to be single.

02225-2349 = HIP 11072 = κ For = TOK 40. The companion was first resolved in 2007 (Tokovinin & Cantarutti 2008). New measurements are roughly compatible with the 26.5-year astrometric orbit of Gontcharov & Kiyaeva (2002) if we adjust the semimajor axis to 065 and the node position angle to 120°. New photometry (Δy = 5.0, ΔHα = 4.3) shows that the companion is not as faint as measured initially, and that it is redder than the primary star. It will be very useful to measure the relative brightness of the companions in the near-IR with adaptive optics.

05086-1810 = HIP 23932 = WSI 72. Speculated to be a close binary by Henry et al. (2002), it was first resolved in 2006 with the USNO speckle camera on the Blanco 4 m by B. D. Mason et al. (2010, in preparation) at about the same position angle and twice the separation measured here.

05354-0555 = HIP 26241 = CHR 250Aa,Ab. New measurements confirm slow rectilinear motion of this pair due either to a long-period orbit or to Ab being an unrelated background star, in agreement with the conclusions of Mason et al. (2009). Our photometry indicates that CHR 250Ab is bright, V∼6^m.

06003-3102 = HIP 28442C = GJ 225.2C = TOK 9CE. The E companion in this nearby quadruple system was discovered with adaptive optics in 2004 (Tokovinin et al. 2005). In that paper, a tentative astrometric orbit with 23.7-year period was proposed, and an unusually "blue" J − K color of E was noted. The component E was marginally resolved in the visible during the first speckle run at SOAR (Tokovinin & Cantarutti 2008). In the present data, the companion is seen reliably above the detection limit, and its magnitude difference is measured consistently as Δy = 4.5^m. All four position measurements available to date show retrograde orbital motion (Figure 13) which does not follow the preliminary astrometric orbit suggested by Tokovinin et al. (2005), but is still compatible with a 24-year orbital period. Within a few years, the orbit can be established more firmly and we will be able to address the properties of this apparently peculiar companion. New observations of the other sub-system AB confirm its orbital elements.

06410+0954 = HIP 31978 = 15 Mon = CHR 168Aa,Ab. The most recent orbit of Gies et al. (1997) is clearly in error. This system has been the subject of regular observation by both speckle interferometry and HST–FGS. A new orbit is currently in preparation.

Zoom In Zoom Out Reset image size

07523-2626 = HIP 3840 = V402 Pup = WSI 54. This 9.23^m star of spectral type O6e belongs to the open cluster NGC 2467. Mason et al. (2009) have resolved it into a close pair WSI 54 and measured in 2006.194 the position angle 2318 and the separation 0091. Our observations clearly show three stars in a tight linear configuration (Figure 12). The outer companion matches the WSI 54 pair best, the inner companion has a separation two times smaller and a similar flux. This detection is based on three independent data cubes; the companions are not aligned with the AD, so we are confident that this is not an artifact. Further observations will reveal whether this is a dynamically unstable system (trapezium), an unusual multiple with orbits in resonance, or a chance projection of a binary and a single star, more probable in a cluster than in the field.

17248-5913 = HIP 85216 = I 385 + WSI 85. Quite unexpectedly this star, previously considered as a binary, turned out to be a spectacular triple with components of comparable magnitude and separation (trapezium type); see Figure 12. During the 0.72-year time between the Blanco and SOAR09 runs, the relative position of both companions changed only slightly; the magnitude differences remained stable as well. Comparing our measurements with published data, we identify the wider companion at 122°, 039 with the previously known component B and designate the new companion at 270°, 026 as D. The distant companion C at 210°, 17'' (also seen in the Two Micron All Sky Survey (2MASS) images) is 5^m fainter than A and has moved only slightly since its discovery in 1901. Therefore, C likely belongs to this system.

The AB pair has revolved by 64° over the 110 years since its discovery by Innes (1905), suggesting an orbital period of ∼600 years; this corresponds also to the dynamically estimated period at a distance of 211 pc measured by Hipparcos. The projected separation of AD (55 AU) means that its orbital period should be of the order of 300 years.

The component D was noted in the only previous speckle observation by Hartkopf et al. (1993), but it was not accepted as real at that time. Re-measurement of the correlation peak corresponding to AD yields 0199,  27018 on 1990.3496. It seems that the pair AD is slowly opening up. The AD was not resolved visually at a 1 m telescope by Holden (1977a). The Hipparcos measured the relative position of AB at 118°, 0452 in disagreement with all other data, as though the light center of AD was measured instead of A. The mysterious new companion D deserves further observations.

17535-0355 = HD 162905 = V2610 Oph = TOK 54. According to Pribulla et al. (2009), this is a close quadruple system composed of two double-lined binaries with periods 8.47 and 0.425 days (the latter is also eclipsing) orbiting each other. All components are dwarfs of spectral types F–G. The outer system is resolved here for the first time. The authors estimate Δm = 0.27^m; our measurement Δy = 1^m is biased by the low S/N. Some other eclipsing binaries discovered to be spectroscopic multiples by the same team were also observed at SOAR in 2009 April and found unresolved. However, one of those, HDS 238, has a visual companion at 32 listed in the WDS, too wide to be measured here.

18455+0530 = HIP 92027 = STF 2375AB, FIN 332Aa,Ab & FIN 332Ba,Bb. New orbits, based on new reductions of historical interferometric measures and new measures are in process for this complex multiple system (B. D. Mason et al. 2010, in preparation).

20401-2852 = HD 196718 = SEE 423. This triple system (V = 8.70, F5V) was first resolved in 1897 at (092, 20°). Three observations of this pair are listed in Aitken's (1932) catalog under ADS 14115, showing slow direct motion. Holden (1977b) measured in 1976.8 (076, 32°) and Δm = 0.5. However, a larger separation of (1112, 33°) was measured by the Tycho experiment on 1991.68 (ESA 1997). One year later, in 1992.4552, Hartkopf et al. (1996) found the pair at a very different position, (034, 271°). We see now that this system is a visual triple (Figure 12), with the wide pair AB at (115, 38°) matching the Tycho result and corresponding to SEE 423. The closer pair BC is fainter than A by 0.64^m (Tycho) and was apparently measured for the first time in 1992 by Hartkopf et al. (1996). The triple nature of SEE 423 clarifies some, but not all, contradictions in the existing data. The wider (and the brightest) component was not seen by speckle in 1992 as it was outside the field of view. However, why did Holden and other visual observers not resolved the sub-system BC with nearly equal components? Also, why is the measured separation of the wide pair so discordant, ranging from 076 (Holden) to 115 (this work)?