SCExAO/CHARIS Near-infrared Direct Imaging, Spectroscopy, and Forward-Modeling of κ And b: A Likely Young, Low-gravity Superjovian Companion

SCExAO targeted κ And on UT 2017 September 8 with the CHARIS IFS operating in low-resolution (R ∼ 20), broadband (1.13–2.39 μm) mode (Peters et al. 2012; Groff et al. 2014, 2015). SCExAO/CHARIS data were acquired in pupil tracking/angular differential imaging (ADI) mode (Marois et al. 2006) with the star's light blocked by the Lyot coronagraph with the 217 mas diameter occulting spot. Satellite spots—diffractive attenuated copies of the stellar point-spread function (PSF)—were generated by applying a 25 nm amplitude modulation to the deformable mirror (Jovanovic et al. 2015b). Exposures consisted of 42 coadded 20.6 s frames covering a modest total parallactic angle rotation of ∼105. The data were taken under good, "slow" seeing conditions: 04–05 in the V band with 2–4 m s⁻¹ winds. The real-time adaptive optics (AO) telemetry monitor recorded the residual wavefront error after SCExAO's correction, implying typical exposure-averaged H-band Strehl ratios of 90%–92%.

We used the CHARIS Data Reduction Pipeline (CHARIS DRP; Brandt et al. 2017) to convert raw CHARIS data into data cubes consisting 22 image slices spanning wavelengths from 1.1 to 2.4 μm. Calibration data provided a wavelength solution; using the the robust "least squares" method described by Brandt et al., we extracted CHARIS data cubes. Contemporaneous Keck/NIRC2 observations of HD 1160 calibrated CHARIS astrometry, yielding a spaxel scale of 00162, a field of view of radius ρ ∼ 105, and north position angle offset of −22 (see Appendix A).

Basic image processing steps—e.g., image registration, sky subtraction—were carried out using our CHARIS IDL-based data reduction pipeline, which will later be released alongside a future release of the Python-based CHARIS DRP (i.e., the "CHARIS Post-Processing Pipeline"), and were described in recent SCExAO/CHARIS science/instrumentation studies (Currie et al. 2018; Goebel et al. 2018). Inspection of the data cubes revealed little residual atmospheric dispersion and exposure-to-exposure motion of the centroid position; the spot modulation amplitude translated into a channel-dependent spot extinction of atten_λ = 2.72 × 10⁻³ × (λ/1.55 μm)⁻².

To spectrophotometrically calibrate each data cube, we considered both stellar atmosphere models and the widely used library of Pickles (1998) adopted in our previous CHARIS papers (Currie et al. 2018; Goebel et al. 2018), in the GPI Data Reduction Pipeline (Perrin et al. 2014), and in P1640 analysis of κ And b's near-IR spectrum in Hinkley et al. (2013). As described in the Appendix B, for B9V and some other spectral types the library of Pickles lacks direct measurements in the near-IR and instead adopts an extrapolation from shorter wavelengths that would translate into a miscalibrated companion spectrum. As an alternative, we used a Kurucz stellar model atmosphere (Castelli & Kurucz 2004). Parameters were tuned to closely match those determined from interferometry (Jones et al. 2016): T_eff = 11,000 K, log(g) = 4.0.²⁹

As shown in Figure 1, κ And b is visible in raw CHARIS data, with a peak emission roughly three times (0.5–5 times) that of the local speckle intensity in wavelength-collapsed images (individual channels). In the H band, the companion is about as well separated from the speckle halo as it was in earlier SCExAO/HiCIAO data from Fall 2016 obtained with the vortex coronagraph shown in Figure 6(a) of Kuhn et al. (2018). Inspection of our raw broadband images shows that κ And b would be marginally visible without processing at smaller separations down to ρ ∼ 05.

Zoom In Zoom Out Reset image size

To further suppress the stellar halo and yield a detection of κ And b with a high signal-to-noise ratio (S/N) in each channel, we employed advanced PSF subtraction techniques. We performed PSF subtraction using Adaptive, Locally Optimized Combination of Images (A-LOCI Currie et al. 2012)—a derivative of the LOCI algorithm (Lafrenière et al. 2007).³⁰ In this approach, the PSF I of image slice i in an annular region s is subtracted from a weighted linear combination of other image slice regions {j} in the sequence

$$\begin{eqnarray}&&{{ \mathcal R }}_{i,s}={I}_{i,s}-\displaystyle \sum _{j}{\alpha }_{{ij},s}{I}_{j,s}.\end{eqnarray} \tag{ 1 }$$

In LOCI, the coefficients α_ij,s are determined by solving a system of linear equations that minimize the residuals between the target slice and references in an "optimization" region o, the solution to the linear system

$$\begin{eqnarray}&&{\boldsymbol{A}}\cdot {\boldsymbol{\alpha }}={\boldsymbol{b}},\end{eqnarray} \tag{ 2 }$$

where the covariance matrix ${\boldsymbol{A}}$ and column matrix ${\boldsymbol{b}}$ are

$$\begin{eqnarray}&&{{\rm{A}}}_{{jl}}=\displaystyle \sum _{\mathrm{pixels}\,k}{I}_{{jk},o}{I}_{{lk},o}\quad \mathrm{and}\quad {{\rm{b}}}_{j}=\displaystyle \sum _{\mathrm{pixels}\,k}{I}_{{ik},o}{I}_{{jk},o},\end{eqnarray} \tag{ 3 }$$

by a simple matrix inversion. The subtraction zone s is typically a subset of pixels comprising optimization region o. The set of image slices J used to construct a weighted reference PSF is typically defined by those fulfilling a rotational gap criterion, where a point source in region s has moved some fraction of a PSF footprint, δ × θ_FWHM, between frames i and j due to parallactic angle motion.

In A-LOCI, this approach is modified in several ways. First, it optionally removes pixels within the subtraction zone s from the optimization zone o, which increase point-source throughput and—as shown in Appendix C—makes algorithm forward-modeling more tractable ("local masking"/"a moving pixel mask"; Marois et al. 2010b; Currie et al. 2012, 2015b). Second, it redefines the covariance matrix ${\boldsymbol{A}}$ and column matrix ${\boldsymbol{b}}$, selecting the n image slices that are best correlated with the target image slice over each region o. Third, it rewrites ${\boldsymbol{A}}$ using singular value decomposition (SVD) as ${\boldsymbol{U}}{\boldsymbol{\Sigma }}{\boldsymbol{V}}$, truncating the diagonal matrix, Σ, at singular values greater than some fraction of the maximum singular value (svd_lim) before inverting and thus allowing a low(er)-rank approximation of the covariance matrix ${\boldsymbol{A}}$:

$$\begin{eqnarray}&&{\boldsymbol{\alpha }}={({\boldsymbol{U}}{{\boldsymbol{\Sigma }}}_{\gt {{svd}}_{\mathrm{lim}}}{\boldsymbol{V}})}^{-1}\cdot {\boldsymbol{b}}.\end{eqnarray} \tag{ 4 }$$

We performed two reductions: (1) a conservative one focused on obtaining a high-fidelity spectrum and (2) an aggressive one that maximizes the achieved contrast in our data. In our first, "conservative" approach, we processed data in annular regions for each wavelength channel independently (ADI only). The annular subtraction zone of depth dr = 10 was masked, a weighted reference PSF was constructed from a 75 PSF footprint "optimization" area exterior to the subtraction zone, and the diagonal terms of the covariance matrix were truncated at svd_lim = 2 × 10⁻⁶ × max(Σ). In our second, "aggressive" approach, we performed an A-LOCI reduction first, utilizing ADI only, and then performed spectral differential imaging (SDI) on the ADI residuals. For the ADI component, we shrunk the rotation gap, optimization area, and SVD cutoff, leaving unmasked the subtraction zone, which was dr = 5 pixels deep . For the SDI component, we scaled each image slice in the ADI-reduced data cube by wavelength and subtracted the residuals with A-LOCI. Instead of an angular gap, we imposed a radial gap of δ = 0.65, masked the subtraction zone, and constructed a weighted reference PSF from pixels at the same separation as the subtraction zone but different angles as in Currie et al. (2017b) from the pseudo-inverse of ${\boldsymbol{A}}$ truncated at svd_lim = 1 × 10⁻⁶ × max(Σ). In all cases, given the limited number of exposures, we did not truncate the reference set by cross-correlation. Finally, we descaled, rotated, and combined the ADI/SDI-subtracted image slices together for a final data cube and final broadband (wavelength-collapsed) image.

To assess and correct for signal loss of κ And b due to processing and thus extract a calibrated spectrum and precise astrometry, we forward-modeled planet spectra through the observing sequence (e.g., Pueyo 2016). Our formalism extends that of Brandt et al. (2013), is detailed in Appendix C, and considers both self-subtraction due to displaced copies of the planet signal weighted by coefficients α_ij and perturbations of these coefficients β_ij due to the planet signal.

Figures 2 and 3 display wavelength-collapsed CHARIS images reduced using "conservative" and "aggressive" PSF subtraction approaches and utilizing ADI only and in conjunction with SDI. The companion κ And b is easily visible at a high S/N (88–210) in the wavelength-collapsed images at a projected separation of ρ ≈ 091 and decisively detected in all channels in all reductions. Except for channel 6 in the most conservative reduction (λ_o = 1.376 μm; S/N ∼ 6.4), the detection significance exceeds 10σ in all channels for all reductions.

Zoom In Zoom Out Reset image size

Zoom In Zoom Out Reset image size

To extract the spectrum for κ And b from the conservative (ADI only) reduction, we defined the signal from aperture photometry with r_ap = 0.5λ/D around the best-estimated position (as determined from the wavelength-collapsed image). We repeated these steps with slight modifications to our algorithm settings to confirm repeatability of the spectrum to a level less than the intrinsic S/N of the detection in each channel. We confirmed that a negative copy of the extracted planet spectrum, when inserted into our sequence prior to processing, fully nulled κ And b in all channels after PSF subtraction.

Figure 4 displays the extracted CHARIS spectrum in units of mJy (left) and (right) compares our spectrum to that from P1640 as extracted in Hinkley et al. (2013) in units of erg s⁻¹ cm^–2 Å⁻¹. The spectrum is fully listed in Appendix D. The CHARIS spectrum shows regions of suppressed flux in between the JHK passbands and a slight suppression beyond 2.3 μm, attributed to water and water/CO absorption in early L dwarfs (e.g., Cushing et al. 2005, 2008; Cruz et al. 2018). The H-band spectrum is characterized by a clear peak at λ ∼ 1.65 μm and steep drop at redder wavelengths; the K-band spectrum exhibits a plateau or slightly rising flux between 2.1 and 2.2 μm.

Zoom In Zoom Out Reset image size

The CHARIS spectrum shows slight differences from that extracted from P1640 over wavelengths where the two overlap (1.1–1.8 μm). The CHARIS spectrum is more peaked in the H band than in the P1640 data at ∼1.65 μm, with significantly lower flux density at 1.7–1.8 μm. Section 7.1 discusses the sources of these differences.

Following Greco & Brandt (2016), we assess the nature of residual noise affecting our extracted spectrum by estimating the spectral covariance at κ And b's location in our final data cube. We divided each channel by the residual noise profile and then computed the cross-correlation between pairs of channels i and j in a 2λ/D wide annulus at κ And b's location, masking pixels within 2λ/D of the companion:

$$\begin{eqnarray}&&{\psi }_{i,j}=\displaystyle \frac{ \langle \bar{{C}_{i}{C}_{j}} \rangle }{\sqrt{ \langle \bar{{C}_{i}^{2}} \rangle \langle \bar{{C}_{j}^{2}} \rangle }}.\end{eqnarray} \tag{ 5 }$$

Figure 5 displays the spectral covariance at the location of κ And b. Except for a few red channels (e.g., 16 and 17 in the K band), the covariance drops for sharply off-diagonal elements. The functional form for the covariance proposed by Greco & Brandt (2016) consists of spatially (ρ) and spectrally (λ) correlated noise with characteristic lengths (σ_ρ and σ_λ) and an uncorrelated term A_δ:

$$\begin{eqnarray}&&{\psi }_{i,j}={A}_{\rho }{e}^{-0.5{(({\lambda }_{i}-{\lambda }_{j})/{\sigma }_{\rho })}^{2}}+{A}_{\lambda }{e}^{-0.5{(({\lambda }_{i}-{\lambda }_{j})/{\sigma }_{\lambda })}^{2}}+{A}_{\delta }.\end{eqnarray} \tag{ 6 }$$

The data are best fit by A_ρ = 0.12, A_λ = 0.05, A_δ = 0.82, σ_ρ =0.65, and σ_λ = 0.24: thus, the residual speckle noise is well suppressed and poorly coupled between different wavelengths. At smaller separations where the rotation gap criterion results in far poorer speckle suppression, the noise is dominated by the correlated components (e.g., at ρ ∼ 045, A_ρ + A_λ = 0.56 and A_δ = 0.44).

Zoom In Zoom Out Reset image size

To estimate broadband photometry for κ And b, we convolve the spectrum with the Maunakea Observatories (MKO) JHK filter functions binned down to the resolution of CHARIS. The companion's apparent magnitude in major MKO passbands is J = 15.84 ± 0.09, H = 15.01 ± 0.07, and K_s = 14.37 ± 0.07. Its J – H and J – K_s colors agree with previous estimates from Carson et al. (2013), Hinkley et al. (2013), and Bonnefoy et al. (2014a). In the photometric system of the Two Micron All Sky Survey (2MASS), its colors are slightly redder (e.g., J_2MASS – K_s,2MASS = 1.52).

Cruz et al. (2018) compute L-dwarf near-infrared spectral average templates, constructed (for each spectral type) from a set of characteristic optically spectral-typed substellar objects. The templates cover L0–L4 and L6–L8 field objects, L0–L1 intermediate-gravity dwarfs (L0–L1β), and low-gravity L0–L4 dwarfs (L0–L4γ). The samples of near-infrared spectra comprising each template show typical variations of the order of ∼5% across J – K; inspection of empirical spectra comprising some templates showed variations in spectral shape at similar levels. Thus, we set a floor to the spectrophotometric uncertainty of 5%.

Table 1 and Figure 7 compare how well κ And b matches each template of Cruz et al. Overall, the L0γ template best fits κ And b's spectrum (${\chi }_{\nu }^{2}$ = 1.22), while the L3 field dwarf template fits marginally and the L0–L1β templates are marginally inconsistent at the 95% confidence limit. When focused more on gravity-sensitive H- and K-band, low-gravity templates L0γ and L1γ fit the best; the L1β and L3 field templates are marginally consistent while the L3γ and L4 field templates are marginally excluded. The agreement with the overall shape of κ And b's spectrum drives the small χ² values for the L0γ and L3 field templates; the shape of both the H- and K-band spectra are clearly better fit by the L0γ template.

Zoom In Zoom Out Reset image size

Table 1.  Fits to Spectral Standards of Cruz et al. (2018)

Spectral Type	Gravity Class	Hcont,CHARIS	H2K, CHARIS	${\chi }_{\nu }^{2}$(total)	${\chi }_{\nu }^{2}$(H + K)
L0	field	0.935	1.050	3.76	3.72
L0	β	0.945	1.032	1.68	2.41
L0	γ	0.971	1.020	1.26	1.40
L1	field	0.912	1.056	2.90	3.55
L1	β	0.926	1.056	1.80	1.71
L1	γ	0.949	1.037	2.84	1.41
L2	field	0.896	1.076	2.03	2.44
L2	γ	0.960	1.009	5.10	3.32
L3	field	0.890	1.075	1.51	1.78
L3	γ	0.947	1.031	3.50	1.91
L4	field	0.867	1.075	2.28	1.86
L4	γ	0.940	1.037	15.32	10.96
L6	field	0.847	1.110	3.36	2.94
L7	field	0.855	1.109	5.92	3.19
L8	field	0.794	1.172	5.00	6.63

Note. The ${\chi }_{\nu }^{2}$ values are calculated assuming 15 degrees of freedom for fitting of the JHK peaks and 10 for just H and K. Entries in bold identify those that fit the data to within the 95% confidence limit.

Download table as:  ASCIITypeset image

Our sample of empirical spectra primarily draws from the Montreal Spectral Library³³ and from the VLT spectral library of Bonnefoy et al. (2014b).³⁴ The Montreal library covers MLT dwarfs with field, intermediate (β), low (γ), and very low (δ) gravities characteristic of old (≳1 Gyr), intermediate aged (∼100 Myr), young (∼10–100 Myr), and very young (<10 Myr) low-mass stars and substellar objects, respectively. The Montreal data draw from multiple sources presenting spectra reduced using multiple instruments, including Gagne et al. (2014, 2015), Robert et al. (2016), Aritgau et al. (2010), Delorme et al. (2012), and Naud et al. (2014). The Bonnefoy library focuses on objects near the M/L transition (M6–L1) having intermediate to (very) low gravities (βγδ) with spectra drawn from a single source (VLT/SINFONI) reduced in a uniform manner. We trimmed our Montreal library sample of objects with very low S/N or those with substantial telluric contamination at the edges of the JHK passbands, leaving 360 objects. Since the library of Bonnefoy et al. (2014b) nominally lists a spectrum normalized in J or HK, we focused only on those objects whose spectra can be relatively calibrated across JHK (12 objects).

Figure 8 displays ${\chi }_{\nu }^{2}$ as a function of spectral type for the JHK and HK restricted fits (top and middle panels) and the JHK unrestricted fit (bottom panel), quantitatively showing how well each empirical spectrum matches κ And b's spectrum. The distribution for the restricted fits shows a clear minimum for L0–L1 spectral types with low surface gravity with one (two) objects formally satisfying the 95% confidence limit for the full JHK (HK) spectrum. In both plots, another 2–3 objects lie just above this limit, all of which are likewise L0–L1 objects with low gravity. For the unrestricted fit, more objects cluster at or below the 95% confidence limit, including the ∼10 Myr old objects UScoCTIO 108B (M9.5γ, Bonnefoy library; Bejar et al. 2008) and 2MASS J12074836-3900043 (L1δ, Montreal library; Gagne et al. 2014). The ${\chi }_{\nu }^{2}$ minimum for the unrestricted fit is broader (M9 to L2–L4), although L0–L1γ objects still dominate the subset of those that fit well.

Zoom In Zoom Out Reset image size

Table 2 lists the best-fitting spectra and their properties from the restricted fits. 2MASS J0141-4633 (Kirkpatrick et al. 2006)—an L0–L1γ dwarf and member of the Tucana–Horologium Association—provides the best fit.^{35 ,36} In general, the sample of best-fitting objects listed in Table 2 is dominated by confirmed and candidate L0–L1γ Tuc–Hor members.

Table 2.  Properties of the Best-fitting Substellar Objects

Name	${\chi }_{\nu }^{2}$	${\chi }_{\nu }^{2}$	SpT	Hcont.	H2K	Assoc.	Age	log(L/L⊙)	Teff	log(g)	Mass
	(total)	(H + K)		Index	Index		(Myr)	(approx.)	(K)	(approx.)	(MJ)
2MASS J0141-4633	1.43	1.81	L0–L1γ	0.962	1.027	Tuc–Hor	${40}_{-19}^{+5}$	−3.58	1899 ± 123	4.1–4.2	13–15
									${1800}_{-100}^{+200}$
2MASS J0120-5200	2.25	2.24	L1γ	1.032	1.049	Tuc–Hor	${40}_{-19}^{+5}$	−3.65	1685 ± 145	4.1–4.2	12.5–14
2MASS J0241-5511	1.83	2.59	L1γ	1.015	1.034	Tuc–Hor	${40}_{-19}^{+5}$	−3.67	1731 ± 151	4.1–4.2	12.5–14
2MASS J0440-5126	1.79	2.64	L0γ	1.003	1.006	Tuc–Hor? (53)	${40}_{-19}^{+5}$?	−3.63?	1600–2000	4.1–4.2?	13–15?
2MASS J2033-5635	1.71	2.44	L0γ	0.945	1.034	Tuc–Hor??a	??	??	1600–2000	??	??
2MASS J2325-0259	1.45	2.03	L1γ	1.040	1.067	AB Dor? (65)	130–200?	−3.80?b	1700–1900	4.7–4.9?	30–40?b
2MASS J2322-6151B	1.94	2.26	L1γ	1.015	1.083	Tuc–Hor	${40}_{-19}^{+5}$	−3.68	1793 ± 50	4.1–4.2	12.5–14

Notes. Spectra for all objects match κ And b's at 99.7% confidence for the JHK restricted fit, the HK restricted fit, and the JHK unrestricted fit. Secure moving-group members are defined from Banyan-Σ as those with >95% probability in a given group. Those with >50% are noted with "?": the Banyan-Σ probability is listed in parentheses. Temperatures are listed from Faherty et al. (2016) (first entry) or Bonnefoy et al. (2014a) (second entry) where available; otherwise, they are estimated from the range in temperatures from Gonzales et al. (2018). If given, luminosities, surface gravities, and masses are calculated assuming the nominal object distance, the K-band bolometric correction from Todorov et al. (2010), and the luminosity evolution models of Baraffe et al. (2003).

^aPreviously identified as a Tuc–Hor member, Banyan-Σ favors a field object (∼75% versus 25%). No parallax is given. Thus its membership and properties depending on distance are noted with "??." ^bMass and luminosity estimated using the "optimal" kinematic distance for moving-group membership.

Download table as:  ASCIITypeset image

To illustrate how κ And b's spectrum best resembles that of a low-gravity L0–L1 dwarf, Figure 9 compares it to 2MASS J0141-4633 (the best-fitting object with small spectrophotometric errors) and a representative set of L0–L1 objects with small errors and different gravity classes. The shape of the H and K spectra of κ And b strongly favors that of a low-gravity object, because the H-band spectrum is far sharper than that of any field object and the red half of the K-band spectrum flatter. All other field and intermediate-gravity L0–L1 dwarfs match κ And b more poorly. Other L0–L1γ dwarfs have χ² values that are still characteristically smaller than L0–L1 field objects (see Figure 8).³⁷

Zoom In Zoom Out Reset image size

We use multiple near-infrared spectral indices to assess the companion's surface gravity: the H-continuum index (H-cont) defined by Slesnick et al. (2004) and the H₂K index described by Canty et al. (2013). The H-cont index is defined from two measurements of the "continuum" flux (λ₁ = 1.470 μm, λ₂ = 1.670 μm) and a measurement of the "line" flux at 1.560 μm:

$$\begin{eqnarray}&&{H}_{\mathrm{cont}}=\left[\displaystyle \frac{{\lambda }_{\mathrm{line}}-{\lambda }_{1}}{{\lambda }_{2}-{\lambda }_{1}}{F}_{{\lambda }_{2}}+\displaystyle \frac{{\lambda }_{2}-{\lambda }_{\mathrm{line}}}{{\lambda }_{2}-{\lambda }_{1}}{F}_{{\lambda }_{1}}\right]{/F}_{\mathrm{line}}.\end{eqnarray} \tag{ 8 }$$

The H₂K index is defined as the flux ratio in two small bandpasses in K: H₂K_ind = F_λ,2.17μm/F_λ,2.24μm.

The wavelengths at which these spectral indices are usually evaluated do not perfectly map onto the wavelengths for each CHARIS channel in low-resolution mode, and the width of the bandpasses (Δλ ∼ 0.02 μm) is smaller than the change in wavelength between adjacent CHARIS channels (Δλ ∼ 0.05 μm). Thus, the spectral indices had to be modified. For H-cont, the change is slight: we defined the "line" flux at channel 10 (λ_line = 1.575 μm) and the continuum at channels 8 and 12 (λ_cont. = 1.471 μm and 1.686 μm). Wavelengths listed by Canty et al. (2013) for the H₂K index are more poorly matched to wavelengths defining the CHARIS low-resolution channels. We therefore defined an approximate H₂K index from averages of adjacent channels 19–20 and 20–21: H₂K = (F_λ=2.139μm + F_λ=2.213μm)/(F_λ=2.213μm + F_{λ = 2.290μm}).

Figure 10 compares the H-cont and H₂K indices for κ And b with those from the Montreal and Bonnefoy libraries. For spectral types of M5 to L6, the typical H-cont indices for field dwarfs range from 1 to 0.85. Indices for young dwarfs of low/intermediate gravity from the Montreal and Bonnefoy samples are systematically 0.05–0.10 dex larger, exhibiting very little overlap with the field. The H₂K index appears best at selecting very young (t < 10 Myr) objects dominating the Bonnefoy sample (see also Gagne et al. 2015). The H₂K indices for young dwarfs of low/intermediate gravity are less well separated from the field than H-cont indices. However, they are still characteristically smaller than for field objects, suggesting that this metric may be used to supplement an assessment of gravity derived from the H-cont index. Combining the two indices together retains a clear separation between nearly all young, low-gravity dwarfs and field objects. Thus, although the low resolution of CHARIS's broadband mode precludes a direct application of standard metrics for gravity in the H and K bands, slightly modified versions of these metrics (especially H-cont) can still identify likely  young, low-gravity objects.

Zoom In Zoom Out Reset image size

The measured gravity-sensitive indices for κ And b—H-cont = 1.070 ± 0.039 and H₂K = 1.055 ± 0.041—suggest a low surface gravity. The H-cont index of κ And b is larger than that for any L0–L1 Montreal or Bonnefoy sample object and most similar to H-cont indices for L0–L1 objects classified as having a low gravity. The H₂K index, which is less diagnostic of surface gravity, is less conclusive since κ And b's value overlaps with the values for both field and low-gravity objects. However, considering both indices together, κ And b still stands out as an object that best resembles a low-gravity object.

Our study clarifies the atmospheric and orbital properties of κ And b, summarized in Table 4. Previous studies analyzing broadband photometry and P1640 spectra (Hinkley et al. 2013; Bonnefoy et al. 2014a) admit a wide range of acceptable spectral types or different answers depending on (a) whether field or low-/intermediate-gravity comparison spectra are used or (b) the wavelength range used for matches with empirical spectra.³⁸ Comparing the CHARIS spectra to both optically anchored spectral templates and spectral libraries shows that κ And b best resembles a young, low-gravity L0–L1 dwarf (L0–L1γ) such as 2MASS J0141-4633. Its H-band spectral shape in particular shows strong evidence for a low surface gravity.

Table 4.  Properties of the κ And System

Parameter	κ And A	κ And b	Reference
Object Properties
d (pc)	50.0 ± 0.1	⋯	1
Age (Myr)	${47}_{-40}^{+27}$	$\approx {40}_{-19}^{+34}$?	2, 6
Mass	2.8 M⊙	≈${13}_{-2}^{+12}\,{M}_{{\rm{J}}}$?	2, 6
log(L/L⊙)	${1.80}_{-0.04}^{+1.7}$	−3.81 ± 0.05	2, 6
Spectral type	B9IV/B9V	L0–L1γ	2, 5, 6
Teff (K)	${11327}_{-44}^{+421}$	1700–2000	2, 6
log(g) (dex)	${4.174}_{-0.012}^{+0.019}$	≈4.0–4.5?	2, 4
Photometry
J (mag)	4.26 ± 0.04	15.84 ± 0.09	2
H (mag)	4.31 ± 0.05	15.01 ± 0.07	2
Ks (mag)	4.32 ± 0.05	14.37 ± 0.07	2
L' (mag)	4.32 ± 0.05	13.12 ± 0.1	3, 4
NB_4.05 (mag)	4.32 ± 0.05	13.0 ± 0.2	4
M' (mag)	4.30 ± 0.06	13.3 ± 0.3	4
Astrometry
UT Date	Data Source	[E, N] (arcsec)
2012 01 01	AO188/HiCIAO	[0.884 ± 0.010, 0.603 ± 0.011]	3
2012 07 08	AO188/HiCIAO	[0.877 ± 0.007, 0.592 ± 0.007]	3
2012 11 03	Keck/NIRC2	[0.846 ± 0.010, 0.584 ± 0.010]	2
2013 08 18	Keck/NIRC2	[0.829 ± 0.010, 0.585 ± 0.010]	2
2017 09 05	SCExAO/CHARIS	[0.710 ± 0.012, 0.576 ± 0.012]	2
2017 12 09	Keck/NIRC2	[0.699 ± 0.010, 0.581 ± 0.010]	2

Note. References: (1) Gaia Collaboration, (2) this work, (3) Carson et al. (2013), (4) Bonnefoy et al. (2014a), (5) Hinkley et al. (2013), (6) Jones et al. (2016). We conservatively assign a positional uncertainty in each coordinate to account for the difference between the apparent and actual positions of the star underneath the coronagraph spot (NIRC2, ∼0.25 pixels) or from a polynomial fit to the apparent centroid positions derived from satellite spots (CHARIS, ∼0.25 pixels), uncertainties in the north position angle and pixel scale (larger for CHARIS), the intrinsic S/N (both), uncertainties in the parallactic angle as recorded in the first header (primarily NIRC2), and uncertainties in the astrometry due to self-subtraction/annealing (both, larger for NIRC2). The age, gravity, and mass are not directly measured, so we denote their estimates with "?."

Download table as:  ASCIITypeset image

A number of factors may explain why our conclusions about κ And b's spectrum show small differences from those presented in Hinkley et al. (2013). Chiefly, the S/N of κ And b's spectrum is substantially higher (S/N_med.,CHARIS ∼ 20.5 versus 5 for P1640), in large part owing to SCExAO's extremely high-fidelity AO correction, resulting in a deep raw contrast. This allowed us to extract a higher-fidelity spectrum and more clearly identify which spectral templates and empirical spectra match κ And b. Furthermore, calibrating the κ And b spectrum from P1640 data is arguably more challenging since it relies on forward-modeling data reduced using SDI only (see Pueyo 2016). The slightly wider, redder bandpass (1.1–2.4 μm versus 0.9–1.8 μm) also probes more of κ And b's spectral energy distribution, also aiding the identification of the companion's best-fit spectral properties. Appendix B identifies an additional possible source of differences from the template spectrum used for spectrophotometric calibration.

While Carson et al. (2013) demonstrated that κ And b is a bound companion, their short (∼0.75 yr) astrometric baseline precluded a detailed understanding of the companion's orbit, admitting a wide range of parameter space (Blunt et al. 2017). Our astrometry establishes a 5 yr baseline and decisively determined κ And b's orbital direction (clockwise). Orbital fits from two separate but complementary codes show that the companion's orbital plane is highly inclined relative to the sky and possibly coplanar with the rotation axis of the star. Its eccentricity is likely substantial. The semimajor axis of κ And b suggests that the companion may orbit at a significantly wider separation than previously thought. The companion's orbit—including inclination and semimajor axis—can be better clarified by including new astrometric measurements and determining solutions assuming observable-based priors (Kosmo O'Neil et al. 2018).

While we reserve a detailed atmospheric modeling analysis of κ And b for a future publication, we can use empirical comparisons to now quantitatively limit its temperature, revisit its age, and estimate its surface gravity and mass. Combining these results with new information on κ And b's orbit allows us to revisit a discussion of its plausible formation mechanisms.

Temperature. A subset of the substellar objects whose spectra best fit κ And b have a temperature derived from atmospheric modeling (Bonnefoy et al. 2014b; Faherty et al. 2016). Conveniently, the best-fitting object—2MASS J0141-4633—was analyzed in Bonnefoy et al. (2014a) using models incorporating cloud/atmospheric dust prescriptions that accurately reproduce young, early L-dwarf spectrophotometry over 1–5 μm (Burrows et al. 2006; Daemgen et al. 2017, T. Currie et al. 2018 in preparation). Bonnefoy et al. (2014a) derive T_eff = ${1800}_{-100}^{+200}$ K. While models utilized to constrain temperature in Faherty et al. (2016) were limiting cases that more poorly fit young, early L dwarfs, the derived temperature estimate for 2MASS J0141-4633 using these models is consistent (1899 K ± 123 K). Temperatures for 2MASS J0120-5200, 2MASS J0241-5511, and 2MASS J2322-6151B (all L1γ) are slightly lower, as expected, and consistent with the range of L0–L1γ temperatures listed in Gonzales et al. (2018). Separately, temperatures for the closest-fitting field spectral type (L3) have a comparable range (1800–1900 K; Stephens et al. 2009). Taken together, we estimate a temperature of 1700–2000 K for κ And b.

Age. While a qualitative assessment of "low gravity" generally means "young," the mapping onto age may not be decisive. Specifically, it is not clear yet how systematically different substellar objects are in gravity class from ∼10 Myr to 40 Myr to 100 Myr, etc., and population studies may identify some overlap.³⁹ Nevertheless, we can use properties of the best-fitting substellar objects coupled with system kinematics and interferometric measurements of the primary to determine whether multiple lines of evidence are consistent with the same likely age of the κ And system.

According to the Bayesian analysis tool for identifying moving-group members, Banyan-Σ (Gagne et al. 2018b), four of the seven objects in Table 2 are bona fide, decisive members of Tuc–Hor (>99.7% membership probability), which has an Li-depletion age of ${40}_{-19}^{+5}$ Myr (Kraus et al. 2014). A fifth is a "likely" member of Tuc–Hor (53% probability) and sixth a possible member (25% probability). The other is a previously identified candidate member of AB Dor (130–200 Myr) (Bell et al. 2015), where previous versions of Banyan (e.g., Banyan-II) estimated a far higher membership probability than does Banyan-Σ. Tuc–Hor is comparable in age to the Columba association (t ≈ 30–40 Myr; Zuckerman et al. 2011; Bell et al. 2015), because the two groups' pre-main sequences (luminosity versus temperature) are nearly identical (Bell et al. 2015). While κ And b's proposed membership in Columba is highly suspect (Hinkley et al. 2013), using new Gaia DR2 astrometry Banyan-Σ still suggests that it is a possible member (20% probability).⁴⁰

Thus, regardless of whether κ Andromedae actually is a member of Columba, properties of both the primary and companion are consistent with what a system coeval with Columba should look like. Considering all lines of evidence together, we favor an age of ${40}_{-19}^{+34}$ Myr, where the upper and lower bounds are equated with the upper bound on age for the primary and the lower bound for most best-fitting comparison spectra, respectively.⁴¹

Gravity. While there are few direct anchors for surface gravity for young substellar objects (see Stassun et al. 2006, 2007; Canty et al. 2013, T. Currie et al. 2018, in preparation), atmosphere/substellar evolution models can help identify plausible values for κ And b. Although a small subset of best-fit models that reproduced the spectrum of 2MASS J0141-4633 in Bonnefoy et al. (2014b) had high surface gravities expected for field objects (log(g) ∼ 5–5.5), most had log(g) = 4.0 ± 0.5. Using the evolutionary models of Baraffe et al. (2003), this object, siblings in Tuc–Hor, and slightly younger (20 Myr old) ones are predicted to have surface gravities of the order of log(g) ∼ 4.1–4.2, while those of comparable temperature near our preferred upper age limit of ∼74 Myr should have log(g) ∼ 4.5. Surface gravities of log(g) ∼ 4–4.5 are therefore supported by a joint consideration of detailed atmosphere modeling of best-fitting spectra and predictions from evolutionary models covering κ And's most plausible age range so far.

Mass. Armed with a revised estimate for κ And b's spectral type, photometry, and the system's distance, we calculate a bolometric luminosity of log(L/L_⊙) = −3.81 ± 0.06 using the bolometric correction obtained by Todorov et al. (2010) for 2MASS J0141-4633.⁴² Luminosities for the best-fitting L dwarfs in Tuc–Hor are comparable to that of κ And b or slightly higher by 0.25 dex (−3.55 to −3.8). As their implied masses are 12–15 M_J, if κ And is coeval with Tuc–Hor then κ And b is likely lower in mass. Considering the full range of favored system ages, κ And b's estimated mass is ${13}_{-2}^{+12}\,{M}_{{\rm{J}}}$ and the companion-to-primary mass ratio is q ∼ ${0.005}_{-0.001}^{+0.005}$.

Formation. Our results provide new information helpful for assessing how κ And b relates to bona fide planets detected by both indirect techniques and direct imaging and to low-mass brown dwarfs. While the companion's mass is near or may even exceed the deuterium-burning limit, the utility and physical basis of this IAU criterion or any other hard upper limit on mass for a "planet" is unclear (Luhman 2008).⁴³ Alternative criteria focusing on the demographics of imaged companions—mass ratio and separation—may more clearly distinguish planets from brown dwarf companions (Kratter et al. 2010; Currie et al. 2011).

While the plausible mass ratios of κ And b are intermediate between those of HR 8799 cde (q ∼ 4.5 × 10⁻³) and ROXs 42Bb (q ∼ 9 × 10⁻³), its orbital separation is likely larger than (almost) all HR 8799 planets, more comparable to HIP 65426 b and ROXs 42Bb (90–150 au; Chauvin et al. 2017; Currie et al. 2014b). Similar to κ And b, no additional companions have been found at smaller separations around HIP 65426 or ROXs 42B (Bryan et al. 2016; Chauvin et al. 2017). Although core accretion struggles to form massive companions in situ beyond 50–100 au, disk instability may yet be a viable mechanism to account for κ And b, HIP 65426 b, and ROXs 42Bb (e.g., Rafikov 2005). At least some protoplanetary disks contain a significant amount of mass at separations of 50–150 au scale that could be (and perhaps have been) converted into massive companions via gravitational instability (e.g., Andrews & Williams 2007; Isella et al. 2016), although direct-imaging surveys show that superjovian-mass planets at these separations are rare (Nielsen et al. 2013; Brandt et al. 2014; Galicher et al. 2016).

Follow-up low-resolution CHARIS spectroscopy in individual passbands (J/H/K; R ∼ 80) could better clarify κ And b's atmospheric properties. Gravity-sensitive indices H-cont and H₂K approximated in this work could be more reliably determined; J-band potassium lines (K_I) could provide a third assessment of the companion's gravity (Allers & Liu 2013). An improved census of substellar objects with ages at or just greater than that of Columba/Tuc–Hor (40–100 Myr) aided by the identification of new moving groups (e.g., Gagne et al. 2018a) could better establish a context for κ And b and how its spectrum compares to the full range of objects with very low, low, or intermediate gravity. Ground-based broadband photometry can bracket CHARIS's coverage and also better probe evidence for clouds and small atmospheric dust, while more precisely constraining the companion's temperature (Currie et al. 2011, 2013; Daemgen et al. 2017). Thermal infrared observations with the James Webb Space Telescope could reveal and help begin to quantify the abundance of CO, CH₄, and CO₂ (Beichman & Green 2018).

Higher-resolution (R ∼ 3000) integral field spectroscopy of κ And b achievable with Keck/OSIRIS and later on the Thirty Meter Telescope with IRIS will provide a significant advance in understanding κ And b's gravity, clouds, chemistry, and perhaps formation (Larkin et al. 2006, 2016; Wright et al. 2014). OSIRIS and IRIS spectra can measure narrow gravity-sensitive lines of iron and sodium (Allers & Liu 2013). Fitting these spectra with sophisticated forward models or analyzing their atmospheric retrievals should also yield estimates for CO, H₂O, CH₄, and perhaps NH₃ abundances from resolved molecular line emission (Barman et al. 2015; Todorov et al. 2016). The carbon-to-oxygen ratio derived from these abundance estimates may provide insights into the formation environment of κ And b and perhaps identification with other directly imaged planets (e.g., Barman et al. 2015).