PIERCING THE GLARE: A DIRECT IMAGING SEARCH FOR PLANETS IN THE SIRIUS SYSTEM^*

One and a half centuries ago, Sirius was found to host a faint binary companion (Bond 1862). The orbital motion of this pair has been monitored ever since, leading to a number of publications that claimed to find periodic perturbations indicative of the presence of an unseen third system component, Sirius C. The most recent analysis predicted a substellar companion in a ∼6.3 year circumstellar orbit around Sirius A (Benest & Duvent 1995, and references therein). While the amplitude of the purported astrometric signal, 56 mas, would suggest a companion mass of 72 M_Jup, these authors imposed an upper limit of ≲50 M_Jup on the basis of system stability considerations. Since no measure of confidence is given, we assume a conservative lower limit of half the measured amplitude, 28 mas, on the basis of their plotted periodograms, resulting in a minimal mass of 36 M_Jup.

This would place Sirius C in the so-called brown dwarf desert, the range of orbital parameter space around stars in which brown dwarfs are found to be scarce (e.g., Marcy & Butler 2000; Grether & Lineweaver 2006). Numerical simulations of the formation and evolution of brown dwarf companions reproduce this scarcity, regardless of the formation process assumed (e.g., as part of the star formation process (Bate & Bonnell 2005), planet formation by core accretion (Mordasini et al. 2009), or gravitational fragmentation (Stamatellos & Whitworth 2009)), thus these rare objects impose important constraints on theory.

However, Sirius' extreme brightness—both apparent and absolute—had long thwarted attempts to verify these claims through direct imaging. Kuchner & Brown (2000) established first constraints on substellar companions at separations of 15–3'' with space-based coronography at 1.02 μm on Hubble Space Telescope NICMOS, whereas Bonnet-Bidaud & Pantin (2008) used ground-based observations assisted by adaptive optics on ESO ADONIS to achieve similar constraints in the range of 3''–10''. These limits left most of the parameter space in which Sirius C was expected unexplored. In this work, we present new results from 4.05 μm observations on Subaru IRCS as well as from M-band observations on MMT/AO Clio, which improve the companion mass constraints by an order of magnitude and extend the coverage down to an inner working angle of 07. With a baseline of 4.3 years among the observations, the chance that a companion be missed in all data sets due to adverse geometric conditions is slim.

Sirius A is an A1V-type star at a distance of 2.64 pc, a mass of 2.02 M_☉, and an age of 225–250 Myr, whereas Sirius B is a white dwarf of 0.98 M_☉ with a cooling age of 124 ± 10 Myr orbiting Sirius A with a 50 year period (e.g., Liebert et al. 2005). The combination of youth and extreme proximity (cf. a median distance of 22 pc in the Gemini Deep planet survey; Lafrenière et al. 2007b) allows us to explore unusually small orbital radii, with planetary-mass detection limits down to separations of 2.5 AU in projection.

The Sirius system was observed with the Subaru IRCS instrument (Kobayashi et al. 2000) on 2011 January 20, using the AO188 adaptive optics system (Hayano et al. 2010). The Br α narrowband filter at 4.05 μm was chosen, since this band has been shown both theoretically and observationally to provide the best combination of point-spread function (PSF) quality and brightness contrast for planet detection in the speckle-dominated regime (Janson et al. 2008, 2009, 2010). The data set consists of 56 frames, each comprising 240 co-adds of 0.1 s exposures, for a total integration time of 22.4 minutes. Although IRCS does not offer Lyot coronography, a mirror plate with several circular holes is available to reject light from the primary star. We used a hole with a projected radius of 03 to avoid excessive saturation; nevertheless, the shoulders of Sirius A's PSF locally saturate out to a radius of ∼06. The plate scale was 20 mas per pixel.

Follow-up observations were taken with Subaru IRCS on 2011 March 12. Using the same observing strategy, 83 frames were taken for 33.2 minutes of total integration time.

Furthermore, we make use of M-band imaging data taken on 2006 December 5, with the Clio 3–5 μm imager (Freed et al. 2004) in conjunction with the adaptive secondary mirror on the 6.5 m MMTO telescope (Brusa et al. 2004). The star was nodded 55 along the long axis of the detector after five images were taken. Each of the 322 images consists of 50 co-added exposures of 209.1 ms length, for a total co-added duration of 56.1 minutes. To avoid variations in the pattern of illumination on the Clio detector, the instrument is fixed in orientation with respect to the telescope, resulting in total field rotation of 246 for our Sirius data. Conditions were photometric and very stable, with no cloud cover observed on the MMTO all sky camera during the night. Seeing was estimated from the open loop images of the AO alignment camera, with visible light values of 0.5 to 0.7 arcseconds. The plate scale was 49 mas per pixel.

All observations were taken in pupil-stabilized image orientation to enable the angular differential imaging (ADI) technique (Marois et al. 2006). We employed the locally optimized combination of images (LOCI) algorithm (Lafrenière et al. 2007a) in order to search for faint point sources in Sirius A's speckle halo. This form of ADI is the most powerful high-contrast imaging method currently available, as evidenced by recent direct detections of substellar companions (e.g., Thalmann et al. 2009; Marois et al. 2010) and even circumstellar disks (Thalmann et al. 2010; Buenzli et al. 2010). We employed the LOCI parameters described as optimal for the test data set in Lafrenière et al. (2007a), with a frame selection criterion of 0.5 FWHM to avoid self-subtraction of companion signals. Given the excellent resulting image quality, we refrained from further parameter fine-tuning. In all data sets, the inner working angle of 07 is due to detector nonlinearity and saturation rather than insufficient field rotation.

3.1. Subaru IRCS Br α Data

The final Br α images after ADI reduction are presented in Figures 1(a)–(f). No obvious point sources around Sirius A are visible. To confirm this numerically, we first convolve the image with a sampling aperture of 4 pixel diameter and then calculate the signal-to-noise (S/N) map by dividing the pixel values in concentric annuli around the star by their standard deviation. As expected, we find no signal above the 5σ level, discounting the locally elevated noise in the shadow of the unused mirror holes.

Zoom In Zoom Out Reset image size

We do find a conspicuous signal in the immediate vicinity of Sirius B in the 2011 January data (Figure 1(c)). However, closer inspection of the signal's behavior in the time series of images reveals it to be locked to the position angle of the pupil rather than the field and identifies it as a ghost. No such signal is seen in the 2011 March data. Apart from this, the environs of Sirius B are background limited outside the first Airy ring.

We use the unsaturated PSF of Sirius B in the Br α images for photometric calibration. Although no prior Br α photometry of Sirius B exists, its overall brightness in the near- to mid-infrared can be predicted by a 25,193 ± 37 K blackbody (Barstow et al. 2005; Skemer & Close 2011). Model spectra by Lejeune et al. (1997) for a range of log g values up to 5 cm s⁻² for T_eff= 25,000 K indicate that while absorption lines are present in the mid-infrared, they exhibit rather small equivalent width. For the Br α absorption line we estimate an equivalent width of 0.7 nm and hence conclude that it does not present a significant deviation from a blackbody.

As an independent confirmation, we calibrate Sirius B's brightness using a brief observation of the A1V-type star HD 40138 as a reference. The result, 9.27 ± 0.17 mag, is consistent with our expected value of 9.17 ± 0.10 mag, which is based on the J-band photometry by Bonnet-Bidaud & Pantin (2008) and stellar spectra by Castelli & Kurucz (2003). This provides further evidence that Sirius B does not have an infrared excess from circumstellar dust as proposed by Bonnet-Bidaud & Pantin (2008).

3.2. MMT/AO Clio M-band Data

Individual images from Clio are beam subtracted, and a sigma clipping routine is used to remove hot pixels deviating by more than 3σ from a 5 × 5 pixel region centered on that pixel. The location of the central star is estimated by smoothing the images with a Gaussian kernel with FWHM 10 pixels and estimating the location of the resultant peak. All the science images are then passed into a LOCI data reduction pipeline (Lafrenière et al. 2007a). The LOCI processing carries out reduction in 4 pixel (02) wide rings starting from an inner radius of 4 pixels (02). Each optimization section extends 20 pixels beyond the outer radius of the current ring, ensuring that the optimization section contains the equivalent area of a minimum of 150 PSF cores. After the LOCI background subtraction is performed, the resultant images are rotated into the frame of the sky before being averaged together to give a final output image.

Again, no significant signal above the 5σ threshold is found. The S/N map is shown in Figure 2.

Zoom In Zoom Out Reset image size

3.3. Contrast and Detectable Companion Mass

The 5σ contrast curves established by the three data sets are presented in Figure 3. The partial self-subtraction of point sources during the LOCI reduction is estimated by inserting artificial 10σ point sources into the raw data across the entire usable range of separations from the star and retrieving their surviving flux after subjecting them to the LOCI pipeline (radial filtering and optimized background subtraction). The plotted curves are corrected for this flux loss and therefore represent effective detectable contrast. The process is described in detail in Lafrenière et al. (2007a, Section 4.3).

Zoom In Zoom Out Reset image size

The COND-based evolutionary models (Allard et al. 2001; Baraffe et al. 2003) are used to convert the M-band brightness contrast into detectable mass. For the Br α data, we adopt an updated version of the method used in Janson et al. (2008). COND model spectra (Allard et al. 2001) are used to re-calculate photometric predictions from COND-based evolutionary models (Baraffe et al. 2003) from L' band to the IRCS Br α band, for masses of 1–20 M_Jup and the age of the Sirius system (250 Myr). The measured brightness contrast as function of angular separation is then translated into detectable mass, using these values.

The Br α data are sensitive to masses in the planetary regime down to the inner working angle of 07 around Sirius A (6–10 M_Jup at 1''and 2–3 M_Jup at 2'') and reach 1.6 M_Jup in the thermal background beyond 4'', which includes the immediate surroundings of Sirius B down to 02. The M-band data achieve comparable performance in a more restricted field of view (<35).

Our high sensitivity to substellar companions around Sirius A allows the first thorough test of the hypothesis of a ≲50 M_Jup body, "Sirius C," in a 6.3 year orbit around Sirius A as proposed by Benest & Duvent (1995). As these authors point out, though, a single negative detection does not prove the non-existence of such an object. Even though the semimajor axis of such an orbit is a = 4.3 AU, corresponding to 16 at the distance of 2.64 pc, both projection effects and eccentricity can leave the companion at an apparent separation below the inner working angle of 07 at the time of observation.

To quantify the likelihood of a false negative result, we systematically calculate the projected orbital ellipses for 6.3 year orbits with eccentricities e = {0, 0.04, ..., 0.96} and arguments of periastron ω = {0°, 10°, ..., 350°}. Since we make use of radial detectable mass curves (Figure 3), we are insensitive to the orbit's overall position angle on sky; thus, no variation of the longitude of the ascending node Ω is necessary. As for the inclination, we explore two scenarios: the coplanar case, where Sirius C shares the orbital plane of the Sirius AB system with an inclination of 1366 and the free case, where all system orientations on the unit sphere are considered equally likely, resulting in a statistical weight of sin i for each inclination i = {0°, 1°, ..., 90°}. The mass of Sirius C is sampled as m = {0.5, 1, ..., 20} M_Jup.

On each ellipse, 1000 possible companion positions at the epoch of our M-band observations are chosen, with uniform spacing in eccentric anomaly. We then calculate the corresponding location at the epoch of the other two data sets. The detectable mass curves are evaluated to determine whether or not the companion would have been detected at ⩾5σ at least once in that configuration. For each ellipse, the detection likelihood is averaged over all companion positions, whereby a statistical weight is assigned to each position to account for the non-uniform evolution of eccentric anomaly with time. The weight is proportional to the time that the companion spends at each position over the course of an orbit, approximated by half the distance to the two adjacent positions in the sequence divided by the local orbital velocity. Finally, the detection likelihoods are averaged over all i and ω to yield the detection completeness for each combination of eccentricity e and companion mass m. The results of this analysis are plotted in Figure 4.

Zoom In Zoom Out Reset image size

We find that coplanar companions down to 6 M_Jup can be excluded at the 5σ level at 100% completeness. The completeness remains above 50% down to 2–3.5 M_Jup, depending on eccentricity. If no constraints are imposed on the inclination, edge-on orbits emerge that can hide Sirius C behind Sirius A's glare. As a result, the completeness values for masses above 12 M_Jup drop to 97%–99% for certain eccentricity ranges and down to 90% for masses of 5–7 M_Jup.

The astrometric signal reported by Benest & Duvent (1995), 56 mas, implies a companion mass of 72 M_Jup. The authors furthermore impose an upper mass limit of ≲50 M_Jup on the basis of system stability considerations. While no lower mass limit is given, we derive a conservative estimate of half that value, 36 M_Jup, from their published periodograms. Given our much lower detection limits, our three combined epochs of high-contrast imaging can therefore decisively reject their Sirius C hypothesis.

Although Benest & Duvent (1995) provide no error estimation, we consider a false alarm the most likely explanation for their results. Precision astrometry is known to suffer from a multitude of systematic errors (e.g., subtle changes in pixel scale and orientation, and differential atmospheric refraction which depends on air mass, parallactic angle, and ambient conditions) and has led to a series of spurious detections in the past (e.g., Pravdo & Shaklan 2009 versus Bean et al. 2010 and Lazorenko et al. 2011; van de Kamp 1969 versus Gatewood & Eichhorn 1973).

Leaving astrometric predictions aside, we can also explore the parameter space for other semimajor axes. Figure 5 shows the completeness as a function of semimajor axis a = {0.25, 0.50,..., 10.0} AU and companion mass m, assuming a flat distribution in eccentricity. Although our data are sensitive to planets down to an inner working angle of 07 and down to 1.6 M_Jup at large separations, the completeness values drop quickly at shorter separations (e.g., 50% for a 10 M_Jup object at a = 2 AU). This coincides with the domain of long-term stable planet orbits predicted by Holman & Wiegert (1999), with a critical semimajor axis a_c = 2.17 AU. Therefore, plenty of parameter space remains for unseen planets around Sirius A. The upcoming next generation of high-contrast instruments, such as SPHERE (Beuzit et al. 2010), will offer smaller inner working angles and better contrast performance, and thus stand a good chance to detect such planets. In particular, due to its extreme proximity and brightness, Sirius is the third most promising target (after α Centauri A and B) for the direct detection of exoplanets in reflected light with the SPHERE ZIMPOL imaging polarimeter (Thalmann et al. 2008).

Zoom In Zoom Out Reset image size

One thing to keep in mind is the fact that Sirius B was originally a ∼5 M_☉ progenitor star that expanded into a supergiant ∼125 Myr ago, with potentially dramatic consequences for the system architecture (Liebert et al. 2005). Accretion of ejected material from Sirius B may have caused planets around Sirius A to gain mass and heat, migrate, or form in the first place as second-generation planets (e.g., Perets 2010). Since these processes would leave the planets hotter and brighter than their unperturbed 250 Myr old counterparts, our detectable planet mass curves in Figure 3 are conservative for such objects.

We present three high-contrast imaging data sets of the Sirius system, all of which reach detection performances in the planetary regime (6–12 M_Jup at 1'', 2–4 M_Jup at 2'', and 1.6 M_Jup beyond 4''). This constitutes an improvement of an order of magnitude in detectable planet mass. Taken up to 4.3 years apart, the observations allow us to refute the existence of a substellar companion with a mass of ≲50 M_Jup in a 6.3 year orbit as predicted from astrometry measurements of the Sirius AB system (Benest & Duvent 1995). For a companion mass above 12 M_Jup, the chances of a triple false negative at a 5σ threshold are 0%–4%, depending on eccentricity. For the special case of coplanar orbits, the probability is 0% down to 6 M_Jup). However, we note that our observations leave open the possibility for Jupiter- and Neptune-sized planets around Sirius A, especially at short angular separations.

Furthermore, we confirm the absence dust around Sirius B by the lack of an infrared excess at 4.05 μm within our precision of 0.17 mag.

We thank David Lafrenière for generously having provided us with the source code for his LOCI algorithm. We are grateful for the privilege of observing from Mauna Kea, which holds great cultural significance for the Hawai'ian indigenous community.

Facilities: Subaru (IRCS, AO188) - Subaru Telescope, MMT (AO, Clio) - MMT at Fred Lawrence Whipple Observatory