A lucky imaging multiplicity study of exoplanet host stars – II Free

Abstract

The vast majority of extrasolar planets are detected by indirect detection methods such as transit monitoring and radial velocity measurements. While these methods are very successful in detecting short-periodic planets, they are mostly blind to wide sub-stellar or even stellar companions on long orbits. In our study, we present high-resolution imaging observations of 60 exoplanet hosts carried out with the lucky imaging instrument AstraLux at the Calar Alto 2.2 m telescope as well as with the new Spectro-Polarimetric High-contrast Exoplanet REsearch (SPHERE) high-resolution adaptive optics imager at the ESO/VLT in the case of a known companion of specific interest. Our goal is to study the influence of stellar multiplicity on the planet formation process. We detected and confirmed four previously unknown stellar companions to the exoplanet hosts HD 197037, HD 217786, Kepler-21 and Kepler-68. In addition, we detected 11 new low-mass stellar companion candidates which must still be confirmed as bound companions. We also provide new astrometric and photometric data points for the recently discovered very close binary systems WASP-76 and HD 2638. Furthermore, we show for the first time that the previously detected stellar companion to the HD 185269 system is a very low mass binary. Finally, we provide precise constraints on additional companions for all observed stars in our sample.

1 INTRODUCTION

We live in a golden age for extrasolar planet discoveries. In the past decade several large radial velocity and transit surveys have discovered more than 1200 systems containing extrasolar planets (exoplanet.eu, as of 2015 July). While these indirect detection methods have been incredibly successful, they have a few inherent biases. In particular, while they are very sensitive to short-period companions (often in the order of days or weeks), they are blind to wide (sub-) stellar companions at several tens or hundreds of au. However, more than 50 per cent of all main-sequence stars in the Galaxy and approximately half of all solar-type stars are actually members of stellar multiple systems (Mathieu et al. 2000; Raghavan et al. 2010). It is thus of great interest to investigate the influence of stellar multiplicity on extrasolar planet formation and orbital evolution.

There have been a large number of theoretical and observational studies that investigated the influence of close and wide stellar companions on the various stages of the planet formation process. It is, for instance, believed that close stellar companions will truncate protoplanetary discs and shorten their dissipation time-scale. This has been observationally confirmed e.g. by Bouwman et al. (2006), who found a significantly reduced number of discs in binary systems in their Spitzer survey of the young η Cha star cluster. Other studies such as Kraus et al. (2012) find that this effect is dependent on the binary separation with significant drops of disc occurrences only observed for systems with separations smaller than ∼40 au.

In addition to the initial conditions and time-scales in the protoplanetary disc, stellar companions might also influence the accretion of planetesimals by exciting higher eccentricities and velocities which might lead to more destructive collisions (see e.g. Kley & Nelson 2007 or Paardekooper, Thébault & Mellema 2008). However, recent studies find that this effect might be mitigated by the gravitational force of sufficiently massive discs (Rafikov 2013).

Finally, stellar companions might have a major influence on the observed semimajor axis, inclination and eccentricity distributions of extrasolar planets. Studies by Fabrycky & Tremaine (2007) and Petrovich (2015) suggest that Kozai–Lidov-type interactions between planets and stellar companions, in combination with tidal friction, might explain some of the observed extreme short period orbits. Other studies (e.g. Naoz et al. 2011) suggest that such interactions could explain very eccentric planet orbits or spin-orbit misalignment. For a comprehensive overview of all these effects we suggest the article by Thebault & Haghighipour (2014).

To study these effects, it is necessary to find the true fraction of multiple stellar systems amongst extrasolar planet host stars. Diffraction- or seeing-limited imaging is a primary tool for this purpose, in particular to find multiple stellar systems with planets in S-type orbits, i.e. the planets orbit one of the stellar components of the system. This orbit configuration accounts for the majority of multiple stellar exoplanet systems (see e.g. Roell et al. 2012).

There have been a number of imaging studies in the past such as Eggenberger et al. (2007), Mugrauer, Neuhäuser & Mazeh (2007), Daemgen et al. (2009), Chauvin et al. (2011), Lillo-Box, Barrado & Bouy (2012), or more recently Dressing et al. (2014), Mugrauer, Ginski & Seeliger (2014), Mugrauer & Ginski (2015) and Wöllert et al. (2015).

In this work, we present the results of our ongoing multiplicity study employing the lucky imaging instrument AstraLux (Hormuth et al. 2008) at the Calar Alto 2.2 m telescope. In particular, we present results for 60 systems obtained between 2011 and 2015. Results prior to that can be found in the first publication of our survey in Ginski et al. (2012). Our targets are stars around which an exoplanet has been detected by radial velocity or transit observations and which have not yet been observed with high-resolution imaging. We further limit our sample to stars within ∼200 pc (with few exceptions) so that we are able to confirm detected companion candidates via common proper motion analysis. In addition to our lucky imaging observations, we complement our study with extreme adaptive optics supported images from the new planet hunting instrument Spectro-Polarimetric High-contrast Exoplanet REsearch (SPHERE) (Beuzit et al. 2008) at the ESO/VLT.

We derive astrometric and photometric data of all detected companion candidates and perform common proper motion analysis for all systems with more than one observation epoch. Finally, we provide detailed detection limits on all observed systems.

2 OBSERVATIONS AND DATA REDUCTION

The observations presented in this study were undertaken between 2011 July and 2015 March with the lucky imaging instrument AstraLux at the Calar Alto Observatory. In addition, we present data for one system which was taken with the new SPHERE planet hunting instrument at the ESO VLT during guaranteed time observations (GTO) in 2015 May.

For our lucky imaging observations, we used short exposures times in the same order as the coherence time of the atmosphere (e.g. Hormuth et al. (2008) measure a speckle coherence times at the Calar Alto of 36 ms). We then recorded a large number of individual images (typically 50 000) of which we only used subsets with the highest Strehl ratio (Strehl 1902) for final combination. The lucky imaging technique is described in detail in e.g. Law, Mackay & Baldwin (2006). All lucky imaging observations were undertaken using the SDSS i filter. The electron multiplying gain of the instrument was adjusted individually for each target to enable high signal to noise without saturating the primary star. We also adjusted the focus of the instrument several times during the night to ensure highest image quality. In our 2011, 2013 and 2014 observations in visitor mode, we used the full field of view of the detector of 24 × 24 arcsec with the shortest possible exposure time of 29.54 ms in frame transfer mode. For the brightest targets, we used shorter integrations times without frame transfer mode and less overall frames due to larger overheads, i.e. significantly increased readout time. In the 2015 observations in service mode the instrument was used in windowed mode, reading only half of the field of view. This enabled shorter exposure times of typically 15.03 ms. Details for each system are given in Table 1.

Data reduction of the lucky imaging data included flat-fielding with sky flats taken during dawn, as well as bias subtraction. Bias frames were taken before each science exposure with the same gain settings as the science target. After flat-fielding and bias subtraction, the Strehl ratio in each image was measured and then only the images with the 10, 5 and 1 per cent best Strehl ratios were aligned and combined, respectively.¹ For the final data reduction, we utilized the native AstraLux pipeline available at Calar Alto (described in detail by Hormuth et al. 2008), as well as our own pipeline for the reduction of lucky imaging data. Our own pipeline was used in all those (few) cases where the Calar Alto pipeline produced no output due to software malfunction. Final images with detected known companions as well as new companion candidates are shown in Figs 1 and 2. We show the 2013 data when available, since it is in general of slightly higher quality than the 2014 data due to better weather conditions (higher coherence time, no clouds). To enhance the contrast between the bright primary stars and the faint companion candidates, we have employed high pass filtering on the images.

In addition, we did use SPHERE's near-infrared camera IRDIS (Dohlen et al. 2008) in dual band imaging mode (Vigan et al. 2010) to image the HD 185269 system in Y, J and H band with broad-band filters on 2015-05-02. The specific interest in this system was triggered by an observed elongation of the companion's point spread function (PSF) in our AstraLux observations. We used the minimal exposure time of 0.84 s without coronagraph and with neutral density filter, which led to only minor saturation of the core of the primary star's PSF in Y and H band, and no saturation in J band. For each filter setting, we took a total of 20 individual exposures for a total integration time of 16.8 s. All individual images in each band were median combined and then flat-fielded and dark subtracted. Since we did not apply a dither pattern in this very short observation sequence, we then used a bad pixel mask (created from flat and dark frames) to eliminate bad pixels. Finally, we combined both images of the dual imaging mode in each band. A resulting combined colour image is shown in Fig. 3.

3 ASTROMETRIC CALIBRATION AND MEASUREMENTS

The most reliable method to determine if individual companion candidates are bound to the systems around which they are discovered is to ascertain if they exhibit the same proper motion as the primary star of the system. For this purpose, we are measuring the separation and relative position angle (PA) of all newly discovered companion candidates relative to the primary star. To ensure that our astrometric measurements can be compared between different observation epochs as well as with measurements done with different instrument, we took astrometric calibration images in each observation epoch. In 2013 and 2014, we used the centre of the globular cluster M 15 for this purpose. In the 2015 observation epoch M 15 was not visible and we imaged three wide binary systems instead (HIP 72508, HIP 80953 and HIP 59585). To calibrate the pixel scale as well as the orientation of the detector, we used as reference HST observations of M 15 that were taken on 2011-10-22 with the Wide Field Camera 3 (WFC3; Kimble et al. 2008). In the case of the binary stars, we used all measurements of the respective systems in the Washington Double Star Catalog (Mason et al. 2001) as reference. We applied a linear fit to these available measurements to correct for the slow orbital motion of these wide binaries. For the calibration using cluster data, we measured individual star positions in our AstraLux image and the HST reference image with idl² starfinder (Diolaiti et al. 2000), which fits a reference PSF to each star position. The reference PSF was created from the data itself. We then used our own cross-correlation routines to identify the same stars in both images. Finally, we calculated separations and relative orientations of each star relative to all other stars. This was done for 92 stars in 2013 and 90 stars in 2014. We then used the known astrometric calibration of the HST reference image to calculate an astrometric solution for each individual measurement. To exclude stars with a strong proper motion or possibly misidentified stars, we employed sigma clipping. The final astrometric solution for the 2013 and 2014 observations is the median of all computed solutions. We give the results in Table 2. The listed uncertainties are the standard deviations of all astrometric solutions.

In the case of the binary stars, we only have two objects in the field of view, thus we could not create a reference PSF from the data. Instead, we are fitting a two-dimensional Gaussian to the star positions. We checked that this approach is valid by comparing similar measurements in the cluster images with the starfinder results. The deviations between the two methods were typically much smaller than the measurement accuracy. The result of the binary calibration is also given in Table 2. We used the weighted average of the three solutions calculated from the individual binary systems. For the uncertainty, we conservatively assumed the largest individual uncertainty that we measured. The uncertainty of the calibration includes the uncertainty of the linear orbital motion fit mentioned earlier. We note that calibrations using binary stars are prone to systematic offsets due to unaccounted for (or underestimated) orbital motion of the systems. We thus caution that the result of the 2015 calibration might still suffer from such an offset.

We have one companion candidate which was already observed in July of 2011 for the first time. In this case, we utilized the astrometric calibration derived by us with cluster and binary data in Ginski et al. (2012).

For the SPHERE/IRDIS data, we used the astrometric solution calculated by the SPHERE consortium for the GTO run in which the data were taken. This astrometric calibration was derived from multiple observations of the globular clusters 47 Tuc and NGC 6380, for which also precise HST reference observations as well as proper motions for individual cluster members are available. There is a small dependence of the pixel scale on the utilized filter; for our Y-band observations we used 12.234 ± 0.029 mas pix⁻¹ and −1.78 ± 0.13 deg, while we used 12.214 ± 0.029 mas pix⁻¹ for the J band, and 12.210 ± 0.029 mas pix⁻¹ for the H band (the detector orientation is not influenced by the filter choice). In addition, IRDIS shows a small anamorphism between the detector x and y direction. This was also determined from observations of the globular cluster 47 Tuc. To correct for this anamorphism, we multiplied the separation in y by a factor of 1.0062. A detailed description of the IRDIS astrometric calibration is given in Maire et al. (2015).

The measurements of the relative positions of companion candidates to the primary stars was also done by fitting a two-dimensional Gaussian to both objects since there were no other objects in the field of view to build a reference PSF. Also, it is problematic to build an average reference PSF from different data sets, since the shape of the PSF will highly depend on the atmospheric conditions and the height of the target above the horizon. To ensure that we obtained a stable fitting result, we repeated the fitting procedure for each object at least 20 times with slightly different starting positions and fitting box sizes. For companion candidates that were separated by less than 2 arcsec from the bright primary stars, we removed the primary stars’ bright halo by high pass filtering before we measured the companion candidates position. All results are listed in Table 3. The given uncertainties are the uncertainties of the Gaussian fitting added in quadrature to the uncertainties of the astrometric calibration. Multiple observation epochs were available for several systems. We discuss these systems in the following in detail and test if the companion candidates are comoving with the primary stars.

3.1 WASP-76

WASP-76 was observed by us only once in August of 2014. We detected a faint companion candidate ∼0.44 arcsec to the south-west of the star. Two months later in October of 2014, the target was observed also with AstraLux by Wöllert & Brandner (2015), who also detected this companion candidate and claim that it is likely a bound companion due to the decreasing likelihood of background objects with decreasing separation. We used their discovery astrometric data point, along with our own astrometric measurement, to determine if it is possible to draw conclusions on the proper motion of the object relative to the primary star. The corresponding diagram is shown in Fig. 4(a). In order to achieve an accurate position measurement of this faint source, we employed high pass filtering on the images to remove the bright halo of the exoplanet host.

Due to the short time baseline of only two months, and the large uncertainties given by Wöllert & Brandner (2015, presumably due to worse weather conditions compared with our own detection), it is not possible to draw firm conclusions on the proper motion of the companion candidate. However, we note that our own measurement is in principle more consistent with the object being a non-moving background source rather than a bound companion. Particularly the 1σ deviation of the two separation measurements could be well explained by parallactic displacement of the primary star relative to a presumably distant background source. Any future measurement with a similar precision as our own measurement of 2014 August will be enough to determine the status of this companion candidate.

3.2 HD 185269

A low-mass companion to the HD 185269 system was discovered by us with AstraLux observations in Ginski et al. (2012) with observations performed between 2008 and 2011. We followed up on this companion in our current study with observations taken in 2013 July and 2014 August. We show the image obtained in the 2013 observation epoch in Fig. 1. In this observation epoch, we observed for the first time that the companion appeared extended in north-east/south-west direction, while the PSF of the primary star showed no such distortion. This prompted us to re-observe this system with SPHERE/IRDIS. The much higher resolution extreme AO images of SPHERE show for the first time that the companion is actually a very low mass binary system itself with two approximately equally bright components (see Fig. 3). In addition to the (unresolved) follow-up astrometry performed with AstraLux, we measured the relative position of each binary component to the primary star in all bands of the SPHERE/IRDIS observation. We used again Gaussian fitting to determine the positions of all objects. The primary star shows a very mild saturation of the innermost 2–3 pixels in Y and H band. We measured its position again multiple times to ensure that we reached a good fit (we fit the flanks of the saturated PSF in this case). Final results are listed in Table 4. In addition, we used our measurements to calculate the weighted average of the position of the Bb component with respect to the Ba component. We arrive at a separation of 123.55 ± 0.44 mas (∼5 au projected separation at a distance of 47.37 ± 1.72 pc; van Leeuwen 2007) and a PA of 214.87 ± 0.21 deg.

Since the SPHERE image confirmed that HD 185269 B is a binary, we re-examined our 2013 AstraLux observation in order to provide an astrometric measurement of the relative binary position. This is useful to determine the orbit of the binary and constrain its mass dynamically in later follow-up studies of the system. Due to the marginally resolved nature of the binary source in our 2013 AstraLux data, Gaussian fitting proved to be difficult. Instead, we used the primary star's PSF as template and fitted it to the two components of HD 185269 B using idl starfinder. This fit yielded a separation of 95.6 ± 2.8 mas and a PA of 221.1 ± 1.3 deg of Bb relative to Ba, as well as separations of 4538 ± 14 mas and 4458 ± 14 mas and PAs of 8.39 ± 0.17 deg and 7.72 ± 0.18 deg of Ba and Bb relative to A. As expected for a system with such small separation, we see strong orbital motion between the 2013 and the 2015 observation epoch. Due to the non-optimal weather conditions in 2014, the companion is not resolved in our 2014 AstraLux observation. At least one additional astrometric measurement is needed to constrain the orbital elements of this binary system.

3.3 HD 43691

HD 43691 was imaged by us once in March of 2015. We detected a companion candidate approximately 4.4 arcsec to the north-east of the exoplanet host star. Since we only have one epoch it is not yet possible to determine if the object is indeed related to the HD 43691 system. However, upon close inspection of the companion candidate's PSF we noticed that it appears extended along an angle of roughly 135 deg. A close-up of the companion candidate's PSF, as well as the primary stars’ PSF, is shown in Fig. 5. We actually see at least two distinct peaks in the PSF (signal-to-noise ratio³ of 5.8 and 5.5, separation of ∼84 mas, i.e. 6.7 au at 80.4 pc), which would indicate that the object itself may be a multiple system. We compared the companion candidate's PSF with the PSF of the primary star to exclude that this is merely an effect caused by the observation conditions. However, the primary star’ PSF appears circular in the centre with a halo that is slightly extended in north–south direction, i.e. we see no indication for an intrinsic smearing of the PSF along the angle seen in the companion candidate. We note that there appears to be a third peak directly north of the south-east component of the companion candidate's PSF. This might indeed be a residual of a north–south extended halo, as seen in the primarie's PSF. The object might hence be a binary or even trinary companion to HD 43691 A. However, further observations are required to confirm that the source is comoving with the primary star and that it is indeed a multiple system itself.

3.4 Kepler-37

Kepler-37 (KOI-245, KIC 8478994) was observed by us only once in August of 2014. In this data set we discovered a wide (∼8.5 arcsec) companion candidate south-south-west of the exoplanet host star. Kepler-37 was previously observed by Lillo-Box, Barrado & Bouy (2014), also using AstraLux at the Calar Alto observatory. In addition, it was targeted by Adams et al. (2012) using ARizona Infrared imager and Echelle Spectrograph (ARIES) at the Multiple Mirror Telescope (MMT) observatory. Both studies do not mention the companion candidate recovered in our own AstraLux image, since they are focusing on close companions within 6 arcsec of the primary star.

Since the object was located at such a relatively large separation, we decided to check the 2MASS (Skrutskie et al. 2006) survey for previous detection. While the object was not listed in the 2MASS point source catalogue, it was visible in the reduced J, H and K images. We extracted the astrometric position from the individual 2MASS images using Richardson–Lucy deconvolution and then averaged the results over all bands. For details on the extraction, we refer to our recent study Mugrauer & Ginski (2015). We find a separation of 8.030 ± 0.138 arcsec and a PA of 202.99 ± 1.85 deg in the 2MASS observation epoch of 1998.47. We used the 2MASS data in combination with our more precise AstraLux measurement to test if the discovered object is comoving with the primary star. The corresponding diagram is shown in Fig. 4(b). Even though the uncertainties of the 2MASS measurement are large compared to our AstraLux measurement, the position of the companion candidate in the 2MASS epoch is consistent within 1σ with a non-moving background object. By comparison, co-motion with the primary can be rejected on the 4σ level. We thus conclude that the object is likely located in the distant background and is not physically associated with the Kepler-37 system.

3.5 Kepler-21

Kepler-21 (KOI-975, KIC 3632418) was observed in July of 2013 and August 2014 with Astralux. A very close companion candidate at approximately 0.8 arcsec was detected south-east of the primary star. In order to get an accurate position measurement of this faint source, we employed high pass filtering on the images to remove the bright halo of the exoplanet host. The resulting measurements were compared with the proper motion of the primary star. The corresponding diagram is shown in Fig. 4(c). Due to the direction of motion of Kepler-21, no significant change in separation would be expected for a comoving object as well as a non-moving background object. However, as can be seen in the diagram, both types of objects diverge in expected PA. Our measurements of the PA of the companion candidate show no significant change in PA, consistent with common proper motion. We can reject the background hypothesis with 4.0σ. We thus conclude that the object that we detected is most likely gravitationally bound to Kepler-21 A and is thus a new low-mass stellar companion in this system.

3.6 Kepler-68

Kepler-68 (KOI-246, KIC 11295426) was imaged by us also in July of 2013 and August of 2014. A wide companion candidate approximately 11 arcsec to the south-east was detected in both observation epochs. Unfortunately Kepler-68 exhibits only a very small proper motion of −10.60 ± 1.60 mas yr⁻¹ in declination and −8.50 ± 1.60 mas yr⁻¹ in right ascension. Thus with only one year of epoch difference it was not possible to assert whether the companion candidate is comoving with the primary star. However, since the companion candidate is located at a wide separation, we checked again the 2MASS catalogue to see if the source had been previously detected. We found that our companion candidate is indeed contained in the 2MASS point source catalogue at a relative position of 10.989 ± 0.085 arcsec and 145.45 ± 0.44 deg. Using this additional observation epoch, we tested the companion for common proper motion with the primary star. The corresponding diagram is shown in Fig. 4(d). While the separation is inconclusive due to the tangential direction of motion of the primary star, we would have expected a significant change in PA of our measurements with respect to the 2MASS epoch if the companion candidate was a non-moving background object. Instead, we find that all measurements are consistent with no change in PA. However, due to the large uncertainties of the 2MASS epoch we can only reject the background hypothesis with 2.1σ. We conclude that, given our data, it seems likely that the companion candidate is indeed bound to the Kepler-68 system, but further observations need to be undertaken to strengthen this conclusion.

3.7 HD 188015

HD 188015 was observed by us with AstraLux in 2013 July and 2014 August. In Fig. 2(i), we show our 2013 observation epoch. A total of six companion candidates are visible in the field of view of AstraLux. The high density of objects in the field of view compared to other systems is not entirely surprising since HD 188015 is located in the direction of the Galactic disc (Galactic latitude of +00| $_{.}^{\circ}$|5428). We note that HD 188015 has a known low-mass stellar companion at ∼13 arcsec and a PA of 85 deg, discovered by Raghavan et al. (2006) in Sloan Digital Sky Survey (SDSS; York et al. 2000) data. This companion is outside of the field of view of our AstraLux observations.

In 2014 August observation conditions were not as favourable as in 2013 with shorter coherence times and thin cloud layers passing through during the observations. Thus only the candidate marked as cc1 was re-detected with high signal to noise. Of the other five companions, four were detected marginally with cc5 being the exception. The marginal detections in 2014 did not allow for fitting of a Gaussian to the companion candidates. Instead, we have determined the centre of light with a simple centroid for those four sources. This led to much larger uncertainties of the 2014 astrometry. We none the less used the 2014 astrometry in combination with the known proper motion of the primary star to determine if one or several of the companion candidates are comoving with the primary star. The corresponding diagrams for cc1 to cc6 (with the exception of cc5) are shown in Fig. 4(e)–Fig. 4(i). For cc1, the available astrometry is more consistent with a background object and we can reject common proper motion with HD 188015 A on the 3σ level. For the remaining companion candidates we cannot reject common proper motion or background hypothesis with any significance. This is caused by the larger uncertainty of the 2014 measurements in combination with the short time baseline of only one year. We note, however, that the astrometry of cc2 and cc4 is more consistent with a distant background object, while the same is not true for cc3 and cc6. The latter two remain completely inconclusive due to their opposite behaviour in separation and PA. We point out that this seemingly mixed behaviour could be caused by a non-zero proper motion of these objects. To gain a better understanding of this system, at least one further observation epoch in good observing conditions is required.

3.8 HD 197037

HD 19037 was observed by us with AstraLux in two epochs in 2013 July and 2014 August. We detected a companion candidate approximately 3.7 arcsec to the south of the primary star. Using the proper motion of the primary star, we calculated the expected position of a non-moving background object in 2013 given the 2014 measurement. The corresponding diagram is shown in Fig. 4(j). The astrometry in both epochs is consistent with no significant change in relative position. We can reject the background hypothesis with 4.8σ in separation and 18.6σ in PA. We conclude that the detected object is comoving with HD 19037 A and is thus most likely a new gravitationally bound stellar companion to the system.

3.9 HD 217786

HD 217786 was observed by us on three different occasions in 2011 and 2013 July, as well as in August of 2014. In all three observation epochs, we detected a companion candidate approximately 2.8 arcsec to the south of the primary star. The proper motion for this system is well determined to be −88.78 ± 0.84 mas yr⁻¹ in declination and −170.13 ± 0.61 mas yr⁻¹ in right ascension (van Leeuwen 2007). We show the astrometric measurements as well as the expected behaviour of a background object in Fig. 4(k). The PA of the companion candidate is not changing significantly with time. However, we detect a small increase in separation. The dashed lines in the diagram show the expected change for a circular edge-on orbit. The data points are consistent with such a change within 1σ. We note that even stronger changes in separation are possible for eccentric orbits. The change in separation is much smaller than what would be expected from a background object and is also showing the wrong direction (for a background object the separation should have decreased from 2011 to 2014). In fact, we can reject the background hypothesis with 42.8σ in separation and 18.9σ in PA. We thus conclude that the discovered object is very likely bound to the system and emerges as new low-mass stellar companion. Due to the small change in separation, but no change in PA, we expect the companion to be in a close to edge-on orbit configuration, but longer astrometric monitoring is required to test this hypothesis.

4 PHOTOMETRIC MEASUREMENTS AND MASS DETERMINATION

To determine the masses of the confirmed companions as well as the possible companion candidates, we performed photometric measurements for all our observation epochs. Since the photometry depends on the gain settings of the detector as well as the observation conditions and height of the target, we did not record a photometric standard star and rather give relative photometric measurements of the companions (and candidates) to their primary stars. While the PSFs of all sources in one image are similar, they are changing with observing conditions and elevation of the targets as well, thus it is not possible to build a reference PSF for photometric measurements from the data. We instead decided to perform aperture photometry on all sources. We used the aperture photometry tool (Laher et al. 2012) for these measurements. The aperture size was adjusted for each image individually to encircle the majority of the flux of the companion candidates. The same aperture size was then used to get the reference measurement from the primary star. In the cases where the faint sources were located within the bright halo of the primary star, care was taken to select a sky aperture close to the companion position to accurately subtract the contribution of the primary to the flux in the aperture. In the case of the primary star, we used sky apertures with large separations from the primary in order to not oversubtract flux due to halo contributions. All results are given in Table 3.

The presented uncertainties take into account statistical uncertainties, which were scaled with a factor of |$\sqrt{2}$| to take into account the increased photometric uncertainty of electron multiplying CCDs. In addition, we consider uncertainties in the differential magnitudes from changing aperture sizes, i.e. if we increase or decrease the aperture radius by up to 2 pixels. These were typically in the order of 0.04 mag and were added in quadrature to the statistical uncertainties.

To convert our photometric measurements to masses we used the BT-SETTL evolutionary models for low-mass stars, brown dwarfs and planets (Allard, Homeier & Freytag 2011). These models take the absolute magnitude and the age of an object as input. To compute the absolute magnitude of our confirmed/possible companions we used the apparent magnitude of the host star in the SDSS i band, as well as the distance of the host star. We then assume that the companions are of the same age as the host star. We summarize these input values for all targets in our survey in Table 5. To get a finer model-grid we interpolated (linearly) between different model ages and star magnitudes. The final masses for all confirmed or possible companion candidates are listed along with their derived absolute magnitude in Table 6. The listed uncertainties for the absolute magnitude include the uncertainty of the apparent magnitude of the host star, as well as the uncertainty in the measured differential magnitude and the uncertainty in the distance of the system. The uncertainties listed for the masses of the objects also account for the uncertainty of the system age. In the following, we compare our photometric measurements and mass determination for a few systems with available literature values.

4.1 HD 2638

For the close stellar companion to HD 2638, we find a differential magnitude of 3.11 ± 0.41 mag in the SDSS i band. Using this measurement along with the age, distance and apparent magnitude of the primary star, we find a mass of 0.425|$^{+0.067}_{-0.095}$| M_⊙ for the companion. The companion was originally discovered by Riddle et al. (2015) using Robo-AO in the optical. They have two measurements in the SDSS i band and find differential magnitudes between primary and companion of 3.39 and 3.19 mag (Riddle et al. 2015; Roberts et al. 2015). They do not provide uncertainties for these measurements. However, given our own uncertainties, both values are within 1σ of our own measurement. To compare our mass result with independent measurements, we used the K_s and J-band photometric measurements of the companion, provided in the characterization paper of the object by Roberts et al. (2015). To calculate a mass range we use again BT-SETTL models. We find an approximate mass range of 0.53 M_⊙–0.45 M_⊙. While this is slightly larger than our own SDSS i-band result, both measurements are consistent within our 1σ uncertainties. The small discrepancy might be explained by a potential oversubtraction of background flux in our SDSS i-band images.

4.2 HAT-P-7

We measure a differential SDSS i-band magnitude between HAT-P-7 A and B of 7.556 ± 0.068 mag. This value is in excellent agreement with the measurement very recently reported in Wöllert et al. (2015), who use the same instrument setup and find a value of 7.58 ± 0.17 mag. Using our differential SDSS i-band magnitude and the system parameters listed in Table 5, we arrive at a mass of 0.205|$^{+0.026}_{-0.021}$| M_⊙. This mass estimate is consistent with the mass range given in the discovery paper by Narita et al. (2010), who find 0.17–0.20 M_⊙ from near-infrared and optical photometry. It also agrees with the more recent mass estimated by Ngo et al. (2015), who find a range of 0.196–0.232 M_⊙, also from near-infrared photometry.

4.3 HD 185269

The photometric measurements in SDSS i band of the 2013 and 2014 AstraLux observation of this system are consistent within 1σ with the previous value published by us in Ginski et al. (2012). Besides the unresolved SDSS i-band photometry, the SPHERE data enabled us to take photometric measurements of the individual components of HD 185269 B. Since we do not have additional sources in the field of view other than the primary and the binary companion, we again used aperture photometry to derive the brightness of the binary components. As mentioned previously the primary star is saturated in Y and H band, thus in these bands we could only measure the brightness difference between the binary components. However, our J-band data are unsaturated, which enabled photometric calibration of the binary measurements with the primary star. We list all our results in Table 7. The given J-band magnitudes are assuming that the neutral density filter is flat across the covered wavelength range.

We used again the BT-SETTL models to convert the J-band measurements into masses of the individual components. For this conversion, we utilized the J-band magnitude of the primary of 5.518 ± 0.027 mag (Cutri et al. 2003) and the most recent age estimate by Pace (2013) of 3.49 ± 0.79 Gyr. The system is located at a distance of 47.37 ± 1.72 pc (van Leeuwen 2007). Given the brightness of the binary components, we calculate masses of 0.165 ± 0.008 M_⊙ and 0.154|$^{+0.009}_{-0.008}$| M_⊙ for the Ba and Bb components, respectively.

To compare these results with our unresolved SDSS i-band measurements, we calculated the flux ratio between the two components in Y band. Given the differential brightness measured in our SPHERE image, the flux ratio between the two components is 0.8. To verify that this flux ratio is consistent with our AstraLux observations in SDSS i band, we used the results obtained from PSF fitting of the two components of HD 185269 B in the 2013 AstraLux data mentioned in Section 3. This PSF fitting yields a flux ratio of 0.78, i.e. consistent with the Y-band results obtained with SPHERE. We calculated the expected apparent SDSS i-band magnitudes for both components to be 14.12 ± 0.10 and 14.36 ± 0.10 mag. From these measurements we derived SDSS i-band masses of 0.18 ± 0.01 and 0.16 ± 0.01 M_⊙ for the Ba and Bb component, respectively. These are consistent with our more precise J-band masses within 1σ.

4.4 WASP-76

For WASP-76, we find a differential magnitude in the SDSS i band of 2.58 ± 0.27 mag. This is very consistent with the value of 2.51 ± 0.25 mag recovered by Wöllert & Brandner (2015), using the same instrument setup. Given our differential magnitude and the age, distance and apparent magnitude of the primary star shown in Table 5, we compute a mass of the object of 0.692|$^{+0.074}_{-0.059}$| M_⊙, assuming that it is indeed gravitationally bound.

4.5 HAT-P-32

In the case of the HAT-P-32 system, we find a differential SDSS i-band magnitude of 5.403 ± 0.057 mag for the stellar companion. Given the age, distance and apparent magnitude of the primary star this translates into a mass of 0.340|$^{+0.048}_{-0.024}$| M_⊙. The mass of HAT-P-32 B was also recently estimated by Ngo et al. (2015), who detected the companion in J, H and K band. They arrived at a mass of 0.393 ± 0.012 M_⊙ for J and K band, and 0.4243 ± 0.0085 M_⊙ for H band. Our mass estimate is lower but marginally consistent within 1σ with their J and K band results. We are deviating from their higher H-band mass by 1.5 σ. However, we want to point out that their two mass estimates also deviate by a similar margin. In principle, it is possible that our slightly lower mass estimate is caused by an overestimation of the background, which is dominated by the bright stellar halo, even though we get consistent photometric results with other studies for sources at even smaller separations, such as WASP-76.

5 DETECTION LIMITS

To guide future observations and enable more sophisticated statistical analysis of the multiplicity ratio of exoplanet hosts, we have derived detection limits at various separations for each of our target stars. For this purpose, we first computed the achievable magnitude difference (contrast) compared to the bright primary star at these separations. We assume that an object is detectable when its signal-to-noise ratio is equal or larger than 5. We then use the peak brightness of the bright primary star as calibration value for the signal. The noise at each separation is determined by averaging over the standard deviation measured in 5 × 5 pixel boxes which are centred on each pixel with the respective separation from the primary star. In Fig. 6, we show the average contrast of all our observations along with the best and worst contrast achieved up to a separation of 5 arcsec, at which we reach the background limit. To convert from these magnitude limits to mass limits, we again utilized the BT-SETTL models as described in Section 4. The input values for this conversion are given in Table 5. The final derived mass limits are given in Table 8. In some cases not all necessary input values were available; we then give only the achievable magnitude limit, which can be used to calculate mass limits at a later time, should all the input values become available. In addition, in a few cases the detectable minimum mass was located outside of our model grid. We then give a lower or upper detection limit based on the closest grid value.

Our detection limits depend mostly on the atmospheric conditions during the observations as well as the brightness and distance of the exoplanet host. Since our sample consists mostly of evolved systems with typical ages in the order of a few Gyr, the dependence of the detectable mass limit on the age is less important. We are on average sensitive to masses down to 0.52 M_⊙ outside of 1 arcsec and down to 0.16 M_⊙ in the background limited region outside of 5 arcsec. These detection limits are comparable to our previous study Ginski et al. (2012) in which we used AstraLux on a similar sample of target systems.

6 DISCUSSION OF THE NEW BOUND STELLAR COMPANIONS

6.1 Kepler-21

Kepler-21 B is located at a projected separation of only 87 au, which might indicate that it should have had a strong influence on the planet formation process. One possible scenario might be that the stellar companion excited high eccentricities in the forming planet causing close encounters with the primary star. The eccentricity could have then been damped by tidal heating which would have left the companion on a very short periodic circular orbit. Such scenarios have been suggested to occur in multiple planetary systems, where multiple objects interact dynamically, e.g. by Rasio & Ford (1996). Given that the system is evolved (∼3.6 Gyr), it is consistent that we would now observe the end product of this interaction.

6.2 Kepler-68

The star Kepler-68 hosts three known planets detected via transit and radial velocity observations by Gilliland et al. (2013). The innermost two of these planets have orbit periods in the order of days and masses in the order of several Earth masses and were detected in transit, while the outer planet d was found in radial velocity data and has a much longer orbit period of ∼1.6 yr (semimajor axis of 1.4 au) and higher mass (m·sin(i) = 0.95 M_Jup). The inner planets appear to be on circular orbits, while the outer planet exhibits an orbit eccentricity of 0.18.

The newly discovered stellar component Kepler-68 B is located at a projected separation of 1485 au. Due to this large separation, the expected period of Kozai–Lidov type resonances is in the order of several Gyr. It seems thus unlikely that the stellar component has a major influence on the dynamics of the inner system via this mechanism. It remains unclear if such a widely separated outer stellar component has a strong influence on the circumstellar disc in the planet formation phase. Following the argument of Kraus et al. (2012), who studied the occurrence rate of circumstellar discs in young binary systems, a major influence of the secondary stellar component is only expected for separations of up to 40 au. If this observational result holds true, then Kepler-68 B is too widely separated to have influenced the circumstellar disc around Kepler-68 A. However, it is in principle also possible that the B component is on a very eccentric orbit. If this is the case then close interactions with the inner planets or the planet forming disc might have happened. In the case of a very eccentric orbit, we would expect to find the stellar companion at a wide separation since it spends the majority of the time there. Further high precision astrometric monitoring combined with statistical orbit analysis might shed some light on the orbit of the B component.

Since the source that we now identified as Kepler-68 B was detected in 2MASS, it was included in the Kepler input catalogue. With its large separation of 11 arcsec the new stellar component is not within the ‘classical’ Kepler PSF. However, Kepler-68 is strongly saturated and shows bleeding. Therefore, the changes in flux can only be seen at the end of bleed columns. Gilliland et al. (2013) showed that Kepler-68 B is located almost precisely in the column direction from Kepler-68. In most of the observing quarters of Kepler, the bleeding encompasses Kepler-68 B and, hence, has to be taken into account for the transit measurements. If the light contribution of B is not considered, systematic errors in the system parameters will arise without changing the quality of the transit fit. From the measured magnitude difference of 6.6 mag (see Table 3), we calculated the amount of contaminating light. If B is a binary that exhibits total eclipses, the transit depth measured by Kepler would be 2300 ppm. This is much higher than the detected transit depth of 346 ppm and 55 ppm for Kepler-68 b and Kepler-68 c, respectively (Gilliland et al. 2013). Therefore a partial or grazing eclipse of B could produce a transit signal. However, as shown before by e.g. Latham et al. (2011) or Fressin et al. (2011), it is very unlikely that an eclipsing background object can mimic a multiple planetary system. Furthermore Gilliland et al. (2013) showed that in Quarter 9 Kepler-68 as well as Kepler-68 B are located between columns in such a way that the bleeding terminates before reaching the latter. In this way they proved that Kepler-68 B cannot be the source of the transit signal. Finally, by applying the BLENDER procedure (Torres et al. 2004; Fressin et al. 2011), Gilliland et al. (2013) could rule out all false positive scenarios involving eclipsing binaries and validate Kapler-68 b,c as planets.

6.3 HD 197037

HD 197037 hosts an m·sin(i) = 0.79 M_Jup planet on an ∼2.8 yr period (semimajor axis of 2.1 au), discovered by Robertson et al. (2012). They found that their best-fitting orbit solution for the planet exhibits an eccentricity of 0.22. They also note that they include in their model a linear trend in radial velocity with a slope of −1.87 ± 0.3 ms⁻¹yr⁻¹, which could be attributed to a long period planet of 0.7 M_Jup and a period of ∼12 yr, or possibly a more distant stellar companion of which they find no further evidence.

To determine if the newly detected stellar companion HD 197037 B can be responsible for this linear trend in the radial velocity, we performed a dedicated Monte Carlo simulation. We fixed the system mass to the combined mass of both stellar components, i.e. 0.34 M_⊙ for B and 1.11 M_⊙ for A (Robertson et al. 2012). We then generated random bound Keplerian orbits which are compatible with our astrometric measurement of HD 197037 B. To somewhat narrow the wide parameter space, we restricted our simulation to orbits with a semimajor axis between 3 and 6 arcsec and times of periastron passage within 2000 yr from our astrometric epoch. We created a total of 15 000 such orbits. We then checked which of these orbits would introduce a slope as measured by Robertson et al. (2012) in their measurement period between the beginning of 2001 and 2012. Out of the 15 000 randomly generated orbits, 1217 orbits fulfill this criterion. In Fig. 7, we show the eccentricities and orbit inclinations of all these orbits. We find that there is no strongly preferred region of the parameter space for an orbit of the B component to produce the measured radial velocity slope. In particular we find orbits for the full range of possible eccentricities. The range of possible inclinations is constrained only by the photometric mass estimate of the B component, i.e. the orbit needs to have a minimum inclination of ∼18 deg to produce the radial velocity signal. From our imaging epochs we cannot yet constrain the orbit of the B component, i.e. it is in principle possible that the B component is in a face-on or close to face-on orbit configuration. However given the large range of orbit solutions of the B component that reproduce the measured linear radial velocity trend, we find it likely that this trend is indeed caused by the stellar B component and not by an additional long period planet.

The non-circular orbit of the existing extrasolar planet around the A component might also be well explained by the new stellar companion if they are caught in mutual Kozai–Lidov type resonances. Given the potential very young age of the system of 0.3±0.3 Gyr (Bonfanti et al. 2015), it may also be possible that the stellar B component was not originally a part of the system but was just caught as the result of a stellar flyby in more recent times. This would then have disrupted the original circular orbit of the planet. However, since HD 197037 is not a known member of a star-forming region or young moving group, such an event would seem rather unlikely.

6.4 HD 217786

Moutou et al. (2011) discovered a long period (∼3.6 yr) planet or brown dwarf with a minimum mass of 13 M_Jup around HD 217786 via radial velocity measurements. They found that the best-fitting orbit solution of the object is very eccentric with e = 0.4 ± 0.05. They do not see long-term radial velocity trends in their data.

The new stellar companion HD 217786 B is located at a projected separation of 155 au. To explore whether the large eccentricity of the planetary companion could be caused by Kozai–Lidov resonances with the stellar companion, we calculated the period of possible Kozai cycles. For this purpose we used the formula provided in Takeda & Rasio (2005). We assumed that the semimajor axis of the orbit of the stellar companion is equal to its projected separation and that the orbit is circular. We get a period of ∼6.2 Myr. This period can be approximately an order of magnitude shorter if the stellar companion is on a significantly eccentric orbit itself. Given the large system age of ∼6.5 Gyr, more than a thousand Kozai cycles could have been complete in principle. It is thus conceivable that the eccentricity of the planetary companion is indeed caused by such interactions with the newly discovered stellar companion. However, this is just one possible scenario to explain the eccentricity of the planet and it strongly depends on the actual orbit of the new stellar companion.

7 SUMMARY

We searched for stellar companions around 60 stars known to harbour extrasolar planets using AstraLux at the Calar Alto observatory. We found previously unknown faint companion candidates within the field of view of our observations around 11 of the observed systems. Of these companion candidates, four, namely Kepler-21 B, Kepler-68 B, HD 197037 B and HD 217786 B, emerged as comoving, and thus in all likelihood gravitationally bound, companions. The candidates detected around HD 188015 and Kepler-37 are more consistent with background objects. For the remaining five systems follow-up lucky imaging observations must still be performed to determine the status of the objects, i.e. if they are comoving with the exoplanet host star. The candidate found next to HD 43691 might be of special interest since it may be a low-mass binary itself.

We also present new photometric and astrometric measurements for the previously known companions to the HD 2638, HAT-P-7, HD 185269, WASP-76 and HAT-P-32 systems. Our SPHERE observations of HD 185269 B showed that the companion is actually a very low mass binary itself, making the system one of only 17 triple systems known to harbour extrasolar planets. Continued astrometric monitoring within the next decade will allow us to determine the dynamical mass of the binary companion.

We note that the previously detected companion candidate to WASP-76 (Wöllert & Brandner 2015) is more consistent with a background source given our new astrometric measurement; however, no final conclusion could be drawn due to the short time baseline between the two observational epochs.

Including the first part of our survey presented in Ginski et al. (2012), we have now studied the multiplicity of 125 known exoplanet systems. In this sample we found so far seven new confirmed binary systems. This includes the new systems reported by us in this work and in Ginski et al. (2012), as well as all systems that were first reported in other studies, but that were unknown at the time of our first epoch observation. This yields a multiplicity rate of only 5.6 per cent in our sample. This is much lower than previous values reported by (Roell et al. 2012, 12 per cent), (Mugrauer et al. 2014, 13 per cent) or (Mugrauer & Ginski 2015, 9 per cent). If most of the unconfirmed companions that we report in this study turn out to be bound companions, the multiplicity rate of our study would increase to 9–10 per cent, which would be in better agreement with previous results. One contributing factor to our lower multiplicity rate might be that the majority of our sample is comprised of planetary systems found via the radial velocity method. Radial velocity surveys routinely exclude known binary systems from their target sample. Thus they introduce an inherent bias towards single star systems. However, the same was in principle true for the studies by Mugrauer et al. (2014) and Mugrauer & Ginski (2015).

If the low multiplicity rate that we recover is indeed caused by a bias introduced by radial velocity surveys, then it would be expected that a higher stellar multiplicity rate is found for transiting planets. Wang et al. (2015) present the results of an adaptive optics imaging search around 138 Kepler planet hosts. They find a stellar multiplicity rate of 8.0 ± 4.0 per cent for multiplanet systems and 6.4 ± 5.8 per cent for single-planet systems and stellar companions with semimajor axes between 100 and 2000 au. These values are in principle consistent with the stellar multiplicity rate of 5.6 per cent that we find, which might indicate that the selection bias of radial velocity surveys has no significant influence on our result. However, from a statistical point of view, considering simple random sampling, our sample size is too small for accurate predictions. If we assume a confidence level of 95 per cent, then our estimated level of accuracy for a population size of 1200 exoplanet systems is only 8.2 per cent. Given that our sample is definitely biased towards single star systems, our actual level of accuracy will be worse than this estimate. To get a reliable estimate of the stellar multiplicity of exoplanet systems with a margin of error on the 5 per cent level, a random sample size of 291 systems is necessary considering the known population of ∼1200 confirmed systems. If we consider a much larger population, i.e. all planetary systems in the Galaxy, then a larger random sample size of 385 systems is needed. These are again lower limits considering the potential biases introduced by exoplanet surveys. We are continuing our multiplicity survey in order to provide a homogeneous observation base for statistical analysis.

We would like to thank the very helpful staff at the Calar Alto Observatory for their assistance in carrying out our observations. In particular, we would like to thank S. Pedraz, G. Bergond and F. Hoyo for organizing and carrying out service mode observations for us. This research is based on observations collected at the German-Spanish Astronomical Centre, Calar Alto, Spain, operated jointly by the Max-Plank-Institut für Astronomie (MPIA), Heidelberg, and the Spanish National Commission for Astronomy. This research also based in part on observations obtained at Paranal Observatory in ESO program 095.C-0273(A). CG and MM thank DFG for support in projects MU2695/13-1, MU2695/14-1, MU2695/15-1, MU2695/16-1, MU2695/18-1, MU2695/20-1, MU2695/22-1, MU2695/23-1. HA acknowledges support from the Millennium Science Initiative (Chilean Ministry of Economy), through grant ‘Nucleus P10-022-F’ and from FONDECYT grant 3150643. MS thanks Piezosystem Jena for financial support. SPHERE is an instrument designed and built by a consortium consisting of IPAG (Grenoble, France), MPIA (Heidelberg, Germany), LAM (Marseille, France), LESIA (Paris, France), Laboratoire Lagrange (Nice, France), INAF - Osservatorio di Padova (Italy), Observatoire de Geneve (Switzerland), ETH Zurich (Switzerland), NOVA (Netherlands), ONERA (France) and ASTRON (Netherlands) in collaboration with ESO. SPHERE was funded by ESO, with additional contributions from CNRS (France), MPIA (Germany), INAF (Italy), FINES (Switzerland) and NOVA (Netherlands). SPHERE also received funding from the European Commission Sixth and Seventh Framework Programmes as part of the Optical Infrared Coordination Network for Astronomy (OPTICON) under grant number RII3-Ct-2004-001566 for FP6 (2004-2008), grant number 226604 for FP7 (2009-2012) and grant number 312430 for FP7 (2013-2016). This research has made use of the SIMBAD data base as well as the VizieR catalogue access tool, operated at CDS, Strasbourg, France. This research has made use of NASA's Astrophysics Data System Bibliographic Services. Finally, CG would like to thank Donna Keeley for language editing of the manuscript.

REFERENCES

SUPPORTING INFORMATION

Additional Supporting Information may be found in the online version of this article:

multiplicity_addendum.pdf

(Supplementary Data).

Please note: Oxford University Press is not responsible for the content or functionality of any supporting materials supplied by the authors. Any queries (other than missing material) should be directed to the corresponding author for the article.