THE EVOLUTION OF THE MULTIPLICITY OF EMBEDDED PROTOSTARS. I. SAMPLE PROPERTIES AND BINARY DETECTIONS^*

Not all of the targets in the sample were observed in the course of our study. Some of the objects in the sample are T Tauri stars, while others are Class 0 objects, or a filament or knot in a cloud. A few of the stars in our sample are examples of what have become known as "transitional" objects, i.e. objects between Class I and T Tauri stars with a spectral index near 0. These sources were observed as they satisfied our selection criteria and since the studies by Haisch et al. (2004) and Duchêne et al. (2004) include such objects. We did not attempt to observe the Class 0 objects since there is typically no flux in the near-IR. In some cases, MSX (Price et al. 2001) observations showed that an IRAS point source was a knot or a filament in a cloud, and not a true point source. Such sources have no near-IR counterpart and were not observed.

Previous studies of the Class I binary frequency (Haisch et al. 2004; Duchêne et al. 2004) searched for binary companions at the K band. We found that the seeing was better and more stable at L' than at the K band, and that reflection nebulae (which tend to have blue colors) are much less of a problem at L'. The bright sky background at L' also made it more difficult to see stars without an IR excess, reducing the effect of background star contamination. We therefore focused our search for binary companions on our L' observations, and used the K-band and H-band observations for additional photometry.

Since we wanted to observe a large number of targets from H through L', we chose telescope/instrument combinations that have this capability in one instrument, have good image quality, and have a suitable plate scale. We used the UH 2.2 m telescope with QUIRC (1024² HgCdTe 1–2.5 μm 3' field of view (FOV), Hodapp et al. 1996), the NASA IRTF with the SpeX guider (512² InSb 1–5 μm 1' FOV, Rayner et al. 2003) and NSFCam2 (2048² Hawaii-2RG 1–5.5 μm 82'' FOV), UKIRT with UIST (1024² InSb 1–5 μm 1' FOV, Ramsay Howat et al. 2004), and Subaru Telescope with CIAO (1024² InSb 1–5 μm 22'' FOV, Murakawa et al. 2004) and IRCS (1024² InSb 1-5 μm 1' FOV, Tokunaga et al. 1998 and Kobayashi et al. 2000). Table 2 lists which telescopes were used on which nights. The majority of our observations used UKIRT and UIST, primarily due to the availability of observing time. Due to UKIRT's north declination limit of +60°, IRTF observations targeted sources north of this limit up to the north declination limit of IRTF (+70°). We used Subaru to observe targets north of this, targets for which we did not get good image quality with IRTF, and to observe targets with adaptive optics (AO).

Dithering was used for all observations in order to remove bad pixels and the detector flat field effects. A 3 × 3 dither pattern was typically used, the size of which was usually 5''–10''and depended on the field of view of the instrument and the availability of guide stars. In the case of L' observations, coadds were used to increase the effective integration time per dither position to ∼20 s to increase observing efficiency. Standard stars that have been observed by UKIRT through the MKO filter set (Simons & Tokunaga 2002; Tokunaga & Simons 2002) were selected from the UKIRT faint standard star list, and were observed for photometric calibration. Furthermore, the instruments we used have MKO filters, and thus all of our observations are in the MKO photometric system.

All data were reduced using the following procedure except in the case of the IRTF data, where the data were first divided by the product of the number of coadds and the number of non-destructive reads. A dark frame was made by averaging together ten individual darks of the same exposure time as the science data. This dark frame was then subtracted from each target frame. To make the sky frame, each dark subtracted frame was scaled to have the same median value, then averaged together using a min–max rejection. The resulting sky frame was then normalized using the median value of the pixel counts. Each dark subtracted (non-scaled) target frame was divided by this normalized sky flat. The median sky value for each frame was subtracted from each frame to set the average background counts in each frame to 0. The images were then aligned and averaged together using an average sigma clipping rejection. In addition, images with better than average resolution were combined into a higher resolution image. This rejects images where the seeing was particularly poor or where there was a guiding error. Since most of the L' "sky" brightness is from the telescope, the procedure we used was not optimal for making a true L' flat field. However, the L' flats we used were effective for removing the detector's flat-field response.

For this project, having the best angular resolution possible was critical. Particular attention was paid on maintaining the best focus possible. In the case of our IRTF observations with the SpeX guider, the image resolution was often limited by aberrations either in the telescope or in the instrument. Aberrations in UIST on UKIRT also occasionally limited our resolution at K, but rarely at L'. We used the 006 plate scale in UIST in order to be able to use a longer exposure time at L' and to better sample the point-spread function (PSF). The resulting 1' FOV also allowed objects within 4500 AU of the target to be in the field of view for the closest targets. The angular resolution distributions at H, K, and L' are presented in Figure 5. The median FWHM was 0609 at the H band, 0543 at the K band, and 0335 at the L' band. The L'-band median FWHM includes our AO observations.

Zoom In Zoom Out Reset image size

The selection of targets in nearby dark clouds naturally selected against nearby bright stars that could be used as AO guide stars. To find sources with a suitable visual guide star, we searched through the USNO-B1.0 catalog (Monet et al. 2003) for stars within 40''of the near-IR source that are brighter than R or I = 16. The objects that we observed with AO are presented in Table 3. To reduce the chance of reflection nebulosity interfering with our search for very close companions, we only observed sources with no resolved nebulosity in our seeing-limited L'-band data.

There are a number of cases where enough of the visible light from the YSO is able to escape the cloud to use the YSO itself as a guide star. This raises the possibility that the AO observed sub-sample is, on average, older and more evolved than the sample as a whole. We used the Kolmogorov–Smirnov test to determine if the AO observed sub-sample is different from the whole sample based on the spectral index and bolometric luminosity distributions. The whole sample and the AO observed sub-sample are not statistically different with regard to spectral index or bolometric luminosity at the 3σ level.

Figure 6 shows a 20'' × 20''(3000 AU to 10⁵ AU, depending on distance) field around each target at L'. For targets where multiple near-IR sources do not fit within this field, more than one field is shown. Each near-IR source is labeled with a number that corresponds to that object's photometry presented in Table 4 if there is more than one object in the field. The inset images show regions of interest in more detail. In some cases, the primary star has been subtracted to better show a companion star in the inset image.

Zoom In Zoom Out Reset image size

Zoom In Zoom Out Reset image size

Zoom In Zoom Out Reset image size

Zoom In Zoom Out Reset image size

Zoom In Zoom Out Reset image size

Zoom In Zoom Out Reset image size

Zoom In Zoom Out Reset image size

Zoom In Zoom Out Reset image size

Zoom In Zoom Out Reset image size

Zoom In Zoom Out Reset image size

Zoom In Zoom Out Reset image size

Zoom In Zoom Out Reset image size

Zoom In Zoom Out Reset image size

Zoom In Zoom Out Reset image size

Zoom In Zoom Out Reset image size

We obtained H-, K-, and L'-band data in order to use the near-IR colors to separate embedded YSOs from foreground or background stars. We used archival Canada–France–Hawaii Telescope (CFHT) Skyprobe data to ensure that the data we used for photometry were taken under photometric conditions, characterized by a stable attenuation measurement near 0 throughout the night. On nights that were non-photometric, the photometry was calibrated using field stars in our photometric data or in 2MASS. If we used 2MASS for H- and K-band photometry and we have our own L' observations, the variability of our targets affects the accuracy of the colors that we derive, since the target may have varied in brightness between the 2MASS observations or between the 2MASS and our observations. We also converted the 2MASS photometry to the MKO system. Aperture photometry was performed using IMEXAMINE in IRAF using five aperture sizes (typically 09, 12, 15, 18, 21) while maintaining the same buffer and sky annulus width (both typically 18) for each aperture size. The same procedure was used for our standard stars. We compared the brightness of the target and the standard using the same aperture size to derive a magnitude estimate for each of the five aperture sizes. We then averaged these five estimates together, taking the standard deviation of these measurements as the accuracy to which we could measure the photometry of that individual source given the quality of the data. We made an airmass correction plot using our standard star data. We used the standard deviation of the standard star photometry from the best-fit linear airmass extinction curve as the lower limit to our photometric errors. This error was combined with the individual measurement error via a Pythagorean sum (root of summed squares) to estimate the total photometric error for each object in each filter (δH, δK, and δL' in Table 4). We used the airmass extinction values in Krisciunas (1987) for our airmass correction.

Visual surveys for binary stars are always sensitive to contamination from background stars. One way to minimize background star contamination is by adopting a contrast limit, such that any star fainter than the limit is not considered as a potential companion since the possibility of such a faint star being background contamination is unacceptably high and the chance of it being a real companion is acceptably low. Haisch et al. (2004) used a contrast limit of ΔK = 4, whereas Reipurth & Zinnecker (1993) adopted a Δz = 5 contrast limit. Duquennoy & Mayor (1991) found that nearly all main-sequence binary stars with a solar-type primary have a mass ratio greater than 10:1. In light of this, we should choose an L'-band contrast limit that allows for all binaries with a mass ratio greater than 10:1. Reipurth & Zinnecker (1993) state that for coeval stars on the Hayashi track, the flux ratio approaches the mass ratio as the wavelength increases, with these ratios being effectively equal at 2.2 μm. Consider a binary system with a mass ratio (and thus a photospheric flux ratio) of 10:1, where only the primary star has an infrared excess. In this case, the primary star's infrared excess can be up to three times greater than its photospheric flux without the observed flux ratio of the binary exceeding 40:1. Thus, a contrast limit of ΔL' = 4 satisfies our criteria for not excluding a significant number of real companions.

The angular resolution of the images, the contrast between the primary star and the companion, and, to a lesser degree, the plate scale of the camera affected how close we were able to detect a companion star. PSF fitting and subtraction was done with our L' data only to reveal very close and faint companions. The most successful PSF model was another field star in the same image. Since the image of the field star and target star were taken simultaneously, the PSFs of the two are nearly identical, and thus the field star is an excellent PSF model. However, this method could only be used rarely since the probability of another bright star being in the field of view is quite low. We usually used stars observed just before and just after the target to be subtracted, and combined these two PSFs into a model PSF for the one to be subtracted. The typical peak counts of the residual after PSF subtraction using this method are about 4% of the PSF's peak counts and are typically found about 1 FWHM from the center of the PSF. There were cases where a star had excessive PSF residuals, either from a poor fit or due to scattered light off of circumstellar material at L'. Scattered light was much less of a problem at L' than at K but is still present, especially very close too the star. Excessive PSF residuals affected how close we could detect fake binary companions (described below), and this is reflected in our inner detection limits.

Knowing when companion stars could have been missed is nearly as important as detecting the companions themselves. Our data are less sensitive to close and faint companion stars. Thus, for each target, we needed to determine the closest separation that a companion of a given contrast could have been found so that we could later correct for our incomplete sensitivity to close and faint companions. To do this, we inserted artificial companions at a range of contrasts (ΔL' = 1, 2, 3, and 4 mag fainter than the primary star) into the PSF-subtracted image of each target star, regardless of whether it has a real binary companion. At each contrast level, we inserted 20 artificial companion stars, one at a time, at a known radius but at a random position angle into the PSF-subtracted image. The image of each artificial binary was viewed for 1 s to ensure that we could easily and confidently find the artificial companion. The artificial companion also had to be easy enough to recover so that, if we were examining real data, we would confidently believe that we had found a companion star. Each artificial binary image was followed by an image of blank sky, also for 1 s, because we found that it was too easy to see the artificial companion "jump" around the image if images of artificial binary companions in different locations were viewed consecutively. If the companion could be recovered at least 19 out of 20 times, then the artificial companion would be inserted at a closer separation and the test repeated until the artificial companion could not be reliably recovered 95% of the time. The inner detection limit is determined to be the closest separation where the artificial companion could be reliably detected at least 95% of time. This test was repeated for each of the four contrast levels mentioned above, and for each individual star.

This method has the disadvantage that we know at what separation to expect artificial companions to be found. However, if we placed artificial companion stars at random separations and random position angles, most of the artificial stars would be inserted at a separation either too close to be recovered, or far enough away to be trivially detected. Even in the case where the artificial star is inserted at a random separation, we are most interested in artificial companions in the separation range where it is possible but difficult to detect the artificial companion. The method we used has the advantages that it quickly identifies the inner separation limit, and it uses the same method used for identifying real binary companions. Table 5 lists the binary systems that we identified. Table 6 lists the inner detection limit for each star at four contrast levels, as well as the outer companion acceptance limit (described below) at each of the four contrast levels.

The purpose of imposing an outer separation limit, beyond which no object would be considered as a companion, is to ensure that all candidate companions are likely to be gravitationally bound companions to the primary star and to help eliminate background star contamination. Duchêne et al. (2004) used an outer limit of 10''(1400 AU at the distance of their targets). They argue that this outer limit is much smaller than the typical size of a typical core in the regions they observed; thus these binaries are likely to have formed from the collapse of the same core or filament. Reipurth & Zinnecker (1993) used an outer limit of 1800 AU. They argue that the typical star-to-star separation in a low-density star-forming region is ∼20,000 AU, and is ∼10,000 AU in a high-density region such as the Trapezium cluster. As such, 1800 AU is an order of magnitude smaller than the typical star-to-star separations for the regions that the targets observed by Reipurth & Zinnecker (1993) are in, and thus they argue that these companions are likely to be gravitationally bound.

There are a handful of well-known common proper motion binary stars with very wide separations that are believed to be gravitationally bound. Perhaps the first star to be recognized as a real binary (versus an optical double) is β Capricorni (Mitchell 1767), which has a projected separation of 9400 AU. Lyrae 1 and 2 have a common proper motion and a projected separation of 13,000 AU (Burnham 1978). While it is rare for a companion to have a separation in excess of 2000 AU, it is possible for such widely separated stars to be gravitationally bound. Furthermore, the mean velocity dispersion of CO gas in the Taurus clouds is 1.4 km s⁻¹, and the observed radial velocity dispersion of Class I protostars is consistent with this value (Covey et al. 2006). At this velocity, it would take 1.7 × 10⁴ years, or roughly the Class 0 life time, to drift 5000 AU. Thus, close but gravitationally unbound stars should be more than 5000 AU apart by the time they are visible in the near-IR as Class I YSOs. We accept companions with projected separation as great as 5000 AU in order to include widely separated companions with confidence that they are likely to be gravitationally bound.

The probability of background star contamination within a projected separation of 5000 AU could exceed 5%, which we consider unacceptably high. This is particularly true in regions near the Galactic center. We used star counts in our L' data to estimate the probability of contamination for each target. We counted all stars in our L' images with near-IR colors consistent with field stars. Since there are many fields with no apparent field stars, we grouped these fields into seven regions of Galactic longitude and latitude to improve the count statistics. Having also derived the L' apparent magnitude distribution for all stars in all fields, we used the star counts in a given region to estimate the density of field stars less than L' = 4 mag fainter than each of the primary stars in that region. We used this density and Equation (1) from Correia et al. (2006) to estimate the radius from the star where the probability of contamination exceeds 5%. This angular radius is

$$\begin{equation} \theta = \sqrt{-{\rm ln}(1-P)/\pi \Sigma } \end{equation} \tag{ 1 }$$

where θ is the angular radius with the probability of contamination P (in our case, P = 0.05), and Σ is the surface density of field stars in that region of Galactic longitude and latitude that are less than L' = 4 mag fainter than the YSO in question. If the 5% contamination radius has a projected radius less than 5000 AU, then the contamination limited radius was used as the outer limit for accepting companions. Otherwise, 5000 AU was used. Although the maximum chance of contamination is 5%, the average chance of contamination within the adopted outer separation limit is 3.0%. Thus we expect there are ∼6 (189 × 0.03) stars identified as binary companions that are background contamination stars. We note that there are a number of fields where the stellar density is so high that we can not confidently identify which object in the field, if any, is the near-IR counterpart to the IRAS source. These targets were thrown out and not considered as having been observed.

The goal of observing our targets in three bands was to use the location of each star observed (target, candidate companion, or background star) on an H − K versus K − L' color–color diagram to minimize the chance of background star contamination. The H − K versus K − L color–color diagram is divided into three main regions: a region of forbidden colors to the left of the reddening vector from the main sequence, a region of colors consistent with a reddened T Tauri star, and a region of colors consistent with a protostar having an IR excess greater than a T Tauri star (the region for reddened main-sequence stars is very narrow). The colors of unreddened T Tauri stars (in the CIT photometric system, not in the MKO system) were adopted from Meyer et al. (1997). The direction of the reddening vector was derived from interstellar extinction values (assuming R = 5, characteristic of dense clouds) taken from Mathis (2000). Using these values, the H − K reddening is 0.079 per magnitude of A(V) extinction, and the K − L (not MKO L') reddening is 0.066 per magnitude of A(V) extinction. Thus, the reddening vector has a slope of 1.20 on the H − K versus K − L color–color diagram.

We excluded those stars whose colors are consistent with a reddened or unreddened main-sequence star or with forbidden colors, with due caution. The colors of a close companion star are difficult to determine accurately due to the proximity of the brighter primary star. Photometric errors and variability affect the measured colors. Reflection nebulosity can strongly affect the observed colors of a star, especially at the H and K bands. Nebulosity makes the star appear bluer, and may not be spatially resolved. The colors of a protostar can range from the forbidden region (if the near-IR flux is dominated by scattered light) to the region characteristic of objects with a strong IR excess. As such, the color information had to be used with other selection criteria, such as the proximity to the IRAS position and the presence of a spatially resolved reflection nebula, to decide if a star is likely to be an embedded YSO or background contamination. Star counts were used along with colors since colors alone are not sufficient to mitigate the chance of background star contamination.

Choosing which candidate companion stars would be retained for further consideration depended on several factors. Stars with H − K and K − L' colors near 0 are likely to be foreground stars and were excluded. An optical or IR reflection nebula is a clear sign that the object in question is physically associated with the cloud. Accurate colors often could not be determined for very close companions. Given the very low probability of contamination at such close separations, these candidate companions were kept.

Figure 7 shows the range of separations and contrasts over which we actually found binary companions. The number of companions versus log(separation/1'') is relatively constant. When plotted against linear separation (arcseconds), most of the binaries have separations less than 3''. Also, for most of the range of separations, we are not less sensitive to fainter companions than brighter ones. It is only within a few times the FWHM (typically less than 1'') that we are less sensitive to faint companions due to the glare of the primary star and, in a few cases, nebulosity. At wider separations, we can be less sensitive to faint companions since our data are not always deep enough to detect companions ΔL' = 4 fainter than the primary star.

Zoom In Zoom Out Reset image size