An Optical/Near-infrared Investigation of HD 100546 b with the Gemini Planet Imager and MagAO

HD 100546 was observed on UT 2016 February 27 (program ID: GS-2015A-Q-501). These data are fundamentally new observations that gave us the capacity to present the first spectrum in the H band. A total of 120 × 60 s IFU images were taken in the H band (1.49–1.79 μm, λ/Δλ ≈ 45) in coronographic mode using ADI. A total of 516 of field rotation (FoV) was obtained over the sequence. Observing conditions were stable and good with an average DIMM seeing of 08 and coherence time of 4 ms. The observing parameters are reported in Table 1. Observations of an argon arc-lamp were acquired at the target elevation immediately prior to the sequence and were combined with reference arcs taken at zenith to calculate flexure compensation (Wolff et al. 2014). The θ₁ Ori field was observed to calibrate the plate scale (14.166 ± 0.007 mas.pixel⁻¹) and the correction to the detector position angle of ADI observations between the north and the vertical axis (01 ± 013, Konopacky et al. 2014).

Table 1.  Observing Log of HD 100546

Instrument	UT-date	Camera	Mode	Filter	Exposure time	Nexp	FoV Rotation	$\langle \mathrm{Airmass}\rangle$	$\langle \varpi \rangle$
					(s)		(deg)		('')
GPI	2016 Feb 27	IFS	Spec.	H	60	120	51.6	1.36	0.87
MagAO	2014 Apr 12	VisAO	SDI	Hα/Cont.	2.27	4939	72	⋯	0.58

Note. The airmass and the DIMM seeing ϖ averaged over the observing sequence when available. "Cont." stands for a filter in the nearby continuum of the Hα line.

Download table as:  ASCIITypeset image

Another set of observations of HD 100546 was obtained on UT 2014 December 17 (GS-2014B-Q-500) but with only 129 of FoV rotation. Therefore, this data set was not used in the present paper.

The raw 2D images were initially processed with the GPI Data Reduction Pipeline v1.3.0 (DRP, Perrin et al. 2014), which performs dark subtraction, bad pixel identification and removal, instrument flexure correction, extraction of spectra to create 3D (x, y, λ) data cubes (Maire et al. 2014), interpolation over a common wavelength axis, and distortion correction as measured with a pinhole mask (Konopacky et al. 2014). The positions of the four satellite spots—attenuated replicas of the central occulted star—at each wavelength in each data cube were also measured with the DRP (Wang et al. 2014). The barycenter of each set of spots was then used to align all the images with the location of the star.

The reduced data cubes were then processed to subtract the stellar PSF. Large-scale structures (e.g., temporal variation of the residual turbulence affecting the stellar halo and background) slowly varying from one frame to another are not fully subtracted by ADI algorithms and must be removed before the PSF subtraction. The data were therefore high-pass filtered. An apodized Fourier filter was used to attenuate, following a Hanning profile, spatial frequencies lower than a parametrized cutoff. Different sizes were tested from 4 pixels up to 16 pixels. To subtract the PSF, three ADI algorithms were applied and the results were compared: the standard cADI (Marois et al. 2006), LOCI (Lafrenière et al. 2007; dr = 5 pixels, N_A = 500  FWHM, 3.6 pixels in the H band, g = 1, and N_δ = 0.75 FWHM), and PCA (Soummer et al. 2012; Amara & Quanz 2012; 1–3 Karhunen–Loève (KL) modes kept, a single region from 3 to 100 pixels in radius). For each algorithm, the residual frames were rotated to align north with the vertical axis and combined with a trimmed mean (10%) in both temporal and spectral directions, resulting in a single image.

The data were also processed using the Reference Differential Imaging (RDI) technique, which is much less susceptible to self-subtraction of disk features, and was not available in the previous works, since we have access to the GPIES reference library. First, the data were processed with pyKLIP (Wang et al. 2015), a python implementation of the KLIP algorithm, to produce a RDI PSF-subtracted broadband image. Second, since the RDI pyKLIP pipeline does not handle spectral data cubes, the data were processed with TLOCI to produce 37 RDI PSF-subtracted spectral images. For the first reduction, a library of reference images was created from all th 6265 individual spectroscopic H band observations obtained with GPI as part of GPIES, at the time when the data were reduced, but only with rejected targets with known disks and/or companions to ensure star subtraction (Draper et al. 2016, M. Millar-Blanchaer et al. 2017, in preparation). Each observation was reduced with the GPI DRP as described previously and collapsed over the wavelengths to populate a homogeneous library of broadband images. HD 100546 data cubes were processed through pyKLIP, using the 1000 most correlated reference images in a 10–70 pixel annulus to avoid the mask edges and the satellite spots. This process produces a broadband image subtracted from the star PSF that can be compared to the ADI-reduced images. For the second reduction, the TLOCI quicklook pipeline was used. The library and the HD 100546 data cubes were spatially scaled to align speckles based on the satellite spot locations in each wavelength channel and flux-normalized with the spots. Each reference library data cube was median-combined to create 426 achromatic reference images. For each frame of HD 100546, the 20 most correlated images of the library were median-combined and subtracted to remove the star PSF. No high-pass filter was applied to preserve large-scale disk structures. The RDI-processed images of HD 100546 were scaled back to the original spatial resolution, and rotated to align the north with the vertical axis. We ended up with a 37-channel RDI-reduced data cube that was used to extract the spectrum of the disk.

Magellan Adaptive Optics observations of HD 100546 were taken on UT 2014 April 12 in better than median conditions (average seeing 058). Total integration time for the image set was 187 minutes, and individual exposures were limited to 2.3 s to minimize saturation effects. The data were taken in ADI mode, and a total of 72° of rotation were achieved (see Table 1).

We utilized the Simultaneous Differential Imaging (SDI) mode of MagAO's visible light camera VisAO to image the disk simultaneously in Hα and in the neighboring continuum as part of the Giant Accreting Protoplanet Survey (GAPlanetS, K. B. Follette et al. 2017, in preparation). Under this mode, the continuum channel serves as a sensitive and simultaneous probe of both the stellar PSF and any scattered-light features.

The SDI data reduction and analysis procedures are described in detail in the companion paper to this study (Follette et al., 2017). Briefly, a custom IDL pipeline performed bias subtraction, flat-field correction, splitting in two channels corresponding to Hα and continuum, and star registration. The PSF in each channel was estimated and removed using pyKLIP with 30-pixel-wide annuli and a movement criterion of 12 pixels, a parameter similar to N_δ in LOCI and defined by the number of pixels a source would move azimuthally and radially due to ADI and SDI. The non-coronagraphic images were saturated out at 8 pixels in radius, so a 10-pixel radius mask was applied to all images. The region near the AO control radius, which is dominated by quasi-static speckles, was also masked. SDI images were created by scaling and subtracting each continuum frame from its Hα counterpart, and these were then processed with pyKLIP. SDI processing serves to remove both PSF artifacts common to both channels and any scattered light structures present in the image; excess emission in the Hα channel is thus preserved.

Contrast curves were also generated. The SDI residual image was convolved with an estimate of the PSF, the noise was measured in annuli of 1 FWHM in width and corrected for small sample statistics (Mawet et al. 2014). Algorithm throughput at the location of any accreting source was computed by injecting fake planets into the raw Hα data sets and processing through SDI and pyKLIP. We measure only the contrast limit for an Hα excess source like LkCa 15 b (Sallum et al. 2015). Therefore, we are only looking for accreting objects considering that any forming planet surrounded by a massive circumstellar disk would be heavily fed by gas. We refer to the paper of Follette et al. (2017) for additional details and to Section 4.3 for a discussion of several hypotheses.

Considering that the planet is still in formation, it is fair to assume that it orbits in the disk plane, with a circular orbit. Eccentricity growth, if any, is expected to occur after the gas dissipation, which has not yet occurred in this system. With the caveat that inward migration is possible (Crida & Morbidelli 2007), the Keplerian motion of HD 100546 b at a deprojected distance of 59 ± 2 au around the 2.4 ± 0.1 M_⊙ central star is 14 ± 01 yr⁻¹, and thus 68 ± 04 projected on the disk plane between the first detection in 2011 June and our observations in 2016 February. Using the position angle of 89 ± 09 from Quanz et al. (2013a), HD 100546 b should be 30 ± 10 northeast of the star in 2016 if rotating clockwise or 148 ± 10 if rotating counterclockwise, since its rotation direction is unknown.

The best-fit astrometry of the source in our GPI data gives a projected separation of 471 ± 9 mas and a position angle of 74 ± 17, where the uncertainties are combined in quadrature from the errors of the measurement, the star center, the plate scale, and the position-angle offset. These measurements are consistent with those of Currie et al. (2015).

All of the reported projected separations remain unchanged to within 1σ over the 4-yr time-span, which is consistent with both a stationary or a Keplerian orbit (only ≈14 mas of total decrease). However, if the H band emission corresponds to a physical object, then its position angle is 1.9σ away from the expected position, assuming clockwise rotation of the system, and 3.7σ away from the counterclockwise prediction.

Therefore, the location of the source seen in our GPI data is consistent with an absence of motion within 1σ. (see Figure 3).

Zoom In Zoom Out Reset image size

Non-Keplerian orbital motion might be expected for HD 100546 b if the planet undergoes migration through disk–planet interaction (Crida & Morbidelli 2007; Baruteau et al. 2016). Given the estimated mass of the planet (several Jupiter masses), classical inward type II migration would be expected on timescales of 10⁴–10⁵ years. However, this scenario predicts a fast decrease of the separation of the protoplanet and a cleared gap (a necessary condition for fast type II migration) at levels that would likely already have been detected (see Section 4.3 for further details on the gap non-detection). The migration rate and direction can be significantly altered if the gap is only partially cleared due to a high viscosity of the disk (Baruteau et al. 2016). Dramatic slowdown of the migration rate or even slow outward migration cannot be rejected based on the astrometry collected so far.

If the source detected in our GPI data is indeed the protoplanet HD 100546 b with an estimated effective temperature of ≃900 K (Quanz et al. 2015), its H band spectrum should be different from that of the surrounding disk. It should exhibit either water absorption if it resembles planets like HR8799 b (e.g., Barman et al. 2011), or a very red slope without molecular absorption if it is very dusty like the hotter 2M1207 b (Patience et al. 2010). Alternatively, if the protoplanet is accreting and the emission mostly originates from the circumplanetary disk, the H band spectrum should have either a red slope or a triangular shape, depending on the accretion rate and the inner disk radius (Zhu 2015). The best-fit contrast of the source in our GPI data is ΔH = 13.2 ± 0.3 mag, where the uncertainties are combined in quadrature with the errors from the measurement and the star-to-spot ratio (Maire et al. 2014). Assuming the source is embedded in the disk, the dust extinction at this location is A_H = 3.4 mag (Currie et al. 2015), leading to H = 15.8 ± 0.3 mag, and $H-L^{\prime} =3.1\pm 0.3$ mag. These values are consistent with the predictions of an accreting protoplanet with ${M}_{{\rm{p}}}\dot{M}=3\times {10}^{-6}\,{M}_{\mathrm{Jup}}^{2}\,{\mathrm{yr}}^{-1}$ and an inner disk radius of 1 R_Jup (Currie et al. 2015). Therefore, the predictions for the source from Zhu (2015) suggest a red featureless H band spectrum.

Our GPI data can test the above scenario by comparing the spectrum of the source with that of the disk. The most aggressive 4-pixel, high-pass-filtered PCA reduction was used for that purpose (Figure 1, left panel), since it shows the resolved source and many bright disk structures. Like the astrometry, the contrast of the source was extracted in each wavelength channel at the best-fit position and FWHM (6.8 pixels). For comparison, contrasts of the disk were also extracted, using aperture photometry with a diameter of 6.8 pixels, from a location southeast of the star at the same projected separation. The regions of these extractions are shown in Figure 4. Since ADI might bias the extracted disk contrasts, they were extracted at the candidate and opposition disk locations in the RDI-reduced residual wavelength channels. All contrasts were then multiplied by the spectrum of the 10,500 K central star (van den Ancker et al. 1997), which was obtained by averaging BT-NextGen models (Allard et al. 2012) at 10,400 and 10,600 K and binning to the resolution of GPI.

Zoom In Zoom Out Reset image size

To estimate the uncertainties of the measurements, each ADI wavelength channel image was convolved by the same aperture. The standard deviation of the pixels at the separation of the extractions was then taken in a region free of astrophysical signal, giving the final error. Typical noise is 14 ± 3% of the source contrast across the band. The same exercise in the RDI-reduced images provides typical 18 ± 2% typical noise.

We do not flux-calibrate the spectra but simply perform a comparison by eye since only the relative slopes and absorption features are of interest. Indeed, calibrating the self-subtraction of the disk due to ADI requires precise modeling (e.g., Esposito et al. 2014) that is beyond the scope of this paper. Nevertheless, since the reduction parameters are the same for all wavelength channels, ADI self-subtraction should be achromatic and only a function of the separation, which is why the measurements were done at the same angular distance.

The final spectra are plotted in Figure 4. The H spectrum of the source at the location of HD 100546 b exhibits no absorption feature and has a blue slope. The comparison with the star spectrum proves the detected emission is pure scattered light, as seen when compared to the disk spectrum at the same separation. The spectrum is also compared to that of several young substellar objects with similar mass in Figure 5: GSC6214-0021 (M9γ, Lachapelle et al. 2015), β Pictoris b (L2γ, Chilcote et al. 2017), HR8799 b (L/Tpec, Barman et al. 2011), and 2M1207 b (L/Tpec, Patience et al. 2010). These spectra are binned to the resolution of GPI in the H band and interpolated over the same wavelength grid. This comparison further emphasizes the dissimilarities with the emission from a young planet.

Zoom In Zoom Out Reset image size

This pure stellar-like spectrum is therefore inconsistent with a planetary photosphere, dust-obscured companion, or accreting circumplanetary disk source.

MagAO Hα images in Figure 6 reveal structure in the inner ∼02 of the disk, and nothing of note outside of this radius. The signal in the inner region is clearly scattered light, as scaling by the stellar Hα/continuum ratio and subtracting creates structure-free SDI images. When these SDI images are KLIP-processed, no Hα excess is apparent at the location of HD 100546 b.

Zoom In Zoom Out Reset image size

Correcting the 5σ ASDI contrast limit at this radius for KLIP throughput reveals that the maximum Hα contrast at the location of HD 100546 b is ∼1 × 10⁻⁴, under the assumption that no flux is present in the continuum channel. The contrast limit at this location can be used to place upper limits on the accretion properties of the protoplanet, given several assumptions as described below. The non-detection of an Hα source at this location is indeed not necessarily inconsistent with the existence of an accreting protoplanet.

Accreting protoplanets, and giant planets in particular, are expected to clear circular regions in the disk, each several Hill radii in extent (about 7 au in our case; Dodson-Robinson & Salyk 2011) and extending azimuthally as they orbit. Therefore, the protoplanet should lie in a region free of extinction by the dust disk and the accretion luminosity should be directly detectable down to the sensitivity limit. Following Close et al. (2014) and Sallum et al. (2015), we use the contrast limit in Hα to derive an upper limit on the accretion luminosity onto the protoplanet. Assuming the extinction toward the protoplanet is the same as the star, we derive A_R = 0.11 mag from A_V = 0.15 mag (Sartori et al. 2003) and the standard interstellar dust extinction law (Cox 2000). Then, given the width and zero-point of the Hα filter, the distance and magnitude of HD 100546, A_R, and the contrast limit, we derive ${L}_{{\rm{H}}\alpha }\lt 1.4\times {10}^{-5}\,{L}_{\odot }$, and considering that accretion laws from T-Tauri stars also apply to planetary-mass objects (Rigliaco et al. 2012), ${L}_{\mathrm{acc}}\lt 1.7\,\times {10}^{-4}\,{L}_{\odot }$. The accretion rate $\dot{M}$ can be computed from the accretion luminosity with the mass (M) and radius of the accreting object (Gullbring et al. 1998). We consider radii of 1 R_Jup and 2 R_Jup to bracket plausible values from evolutionary models (Baraffe et al. 2003), but treat the mass as a free parameter, as it is very uncertain. Therefore, we place upper limits on $M\dot{M}\lt 6.3\times {10}^{-7}\,{M}_{\mathrm{Jup}}^{2}\,{\mathrm{yr}}^{-1}$ and $\lt 1.3\times {10}^{-6}\,{M}_{\mathrm{Jup}}^{2}\,{\mathrm{yr}}^{-1}$, for 1 and 2 R_Jup respectively, under the assumption that the protoplanet is in a clear region. For comparison, with a possible mass from a few to 15 M_Jup for HD 100546 b, the accretion flow on the protoplanet, if any, remains very small in comparison to that on the central star ($M\dot{M}\sim 2\times {10}^{-1}\,{M}_{\mathrm{Jup}}^{2}\,{\mathrm{yr}}^{-1}$, Mendigutía et al. 2015). The upper limit on the accretion rate rules out the previously proposed scenario in which the H band emission is coming from an accreting protoplanetary disk with an inner radius of 1 R_Jup and $M\dot{M}=3.6\times {10}^{-6}\,{M}_{\mathrm{Jup}}^{2}\,{\mathrm{yr}}^{-1}$ (Currie et al. 2015).

The upper limits on the accretion rate and luminosity are typical for T-Tauri stars (Gullbring et al. 1998; Rigliaco et al. 2012), or the young planet LkCa 15 b (Sallum et al. 2015). On the other hand, the protoplanet may be temporarily in a quiet phase and remains undetectable since accretion is known to be stochastic on T-Tauri stars (Bouvier et al. 2007).

Alternatively, it has been hypothesized that HD 100546 b is still embedded in the disk (Quanz et al. 2013a; Currie et al. 2015). The non-detection of any gap larger than a few au around the protoplanet, despite a resolution of 2 au in the optical (Garufi et al. 2016), does not reject the hypothesis of a deeply embedded planet. With A_H = 3.4 mag (Currie et al. 2015), the extinction due to the disk in Hα is A_R = 22 mag. Therefore, the accreting protoplanet would likely not be detectable in Hα. We can still do the same exercise and place some constraints on the accretion luminosity and rate. We find L_acc < 8 × 10³ L_⊙ and $M\dot{M}\lt 1.5\times 10\,{M}_{\mathrm{Jup}}^{2}\,{\mathrm{yr}}^{-1}$ for a 1 R_Jup planet. As expected, these upper limits place no meaningful constraint on the accretion properties of a possible deeply embedded protoplanet.