UNTANGLING THE NEAR-IR SPECTRAL FEATURES IN THE PROTOPLANETARY ENVIRONMENT OF KH 15D

Star B has a K1 spectral type, which was assigned when it re-emerged in 2012 (Capelo et al. 2012). The strongest features we detect from its photosphere are the CO band heads and atomic lines between ∼2.25 and 2.40 μm. Star A, which is a K6/K7 object, should have similar features at these wavelengths despite its slightly cooler temperature (Förster Schreiber 2000). However, we expect that the contribution from star A is negligible in these data. The absorption lines from star B are most prominent in the bright phase spectrum (where the S/N is highest), but stellar features can still be clearly identified in the other three spectra (see Figure 3). The equivalent width measurements from the bright spectrum can be used to verify the spectral classification of star B and constrain the amount of absorption we expect to see at the intermediate and faint phases. We can then remove the stellar contribution from these spectra.

Zoom In Zoom Out Reset image size

Förster Schreiber (2000) showed that absorption line strength ratios in the near-IR are correlated with stellar temperature for cool giants. In order to compare the photospheric absorptions of star B to those seen in other K giants, the equivalent widths of six atomic lines and two CO features were measured (see Figure 4 and Table 4). The line strength quantities with the strongest temperature dependences, as shown by Förster Schreiber (2000), are ${\mathrm{EW}}_{\mathrm{CO}(2.29)}/{\mathrm{EW}}_{\mathrm{Mg}{\rm{I}}}$ and ${\mathrm{EW}}_{\mathrm{Fe}{\rm{I}}(2.2263)}+{\mathrm{EW}}_{\mathrm{Fe}{\rm{I}}(2.2387)}$. The ratios ${\mathrm{logEW}}_{\mathrm{CO}(1.62)}/{\mathrm{EW}}_{\mathrm{Si}{\rm{I}}}$ and ${\mathrm{logEW}}_{\mathrm{CO}(1.62)}/{\mathrm{EW}}_{\mathrm{CO}(2.29)}$ also follow clear relationships with temperature, but the ¹²CO (6–3) line at 1.62 μm was not resolved in our data.

Zoom In Zoom Out Reset image size

Values of ${\mathrm{EW}}_{\mathrm{CO}(2.29)}/{\mathrm{EW}}_{\mathrm{Mg}{\rm{I}}}\,=\,5.7\pm 0.1$ and ${\mathrm{EW}}_{\mathrm{Fe}{\rm{I}}(2.2263)}\,+{\mathrm{EW}}_{\mathrm{Fe}{\rm{I}}(2.2387)}=2.31\pm 0.03\,\mathring{\rm A}$ were obtained from the bright phase spectrum. The measurements indicate a temperature of ∼4000–4500 K, which is consistent with the effective temperature of a ∼K1–K3 giant (van Belle et al. 1999). In addition, the quantity ${\mathrm{EW}}_{\mathrm{Na}{\rm{I}}}+{\mathrm{EW}}_{\mathrm{Ca}{\rm{I}}}\,=4.47\pm 0.06\,\mathring{\rm A}$ and the equivalent width of the ¹²CO (2-0) line at 2.29 μm ($\mathrm{EW}=6.8\pm 0.1$ Å) place KH 15D in between the giant and dwarf branches described by Ishii et al. (2004), as expected for a weak-lined T Tauri star that is contracting to the main sequence. Our data, therefore, confirm the spectral class of star B as around K1, and indicate that it is a sub-giant.

In addition to photospheric absorption, star B shows He i λ10830 emission at all three orbital phases and Paschen α and Brackett γ lines in the bright and intermediate phase spectra. Equivalent widths and absolute line fluxes were measured, and the results are presented in Table 5 (He i) and Table 6 (Paschen α and Brackett γ). Figure 5 shows the He i line profiles from all four spectra, and the hydrogen lines are plotted in Figure 6.

Zoom In Zoom Out Reset image size

Zoom In Zoom Out Reset image size

Table 6.  Equivalent Width Measurements for Atomic Hydrogen Emission

Phase	Feature	Equivalent Width	Absolute Line Flux
	$(\mu {\rm{m}})$	Å	(10−18 W m−2)
Bright	Paschen α	−20 ± 1	7 ± 4
	Brackett γ	−1.17 ± 0.03	0.3 ± 0.1
Intermediate	Paschen α	−30 ± 1	1.3 ± 0.3
	Brackett γ	−1.49 ± 0.05	0.038 ± 0.009
Faint, November	Brackett γ	−1.02 ± 0.03	0.022 ± 0.007

Download table as:  ASCIITypeset image

A redshifted absorption component is observed in the He i feature at the bright phase, which is indicative of gas falling in toward the stars. This inverse P-Cygni shape is additional confirmation that KH 15D is still weakly accreting, although the spectral resolution of our data is not high enough to apply a detailed model of the accretion flow, as has been done with other classical T Tauri stars (e.g., Fischer et al. 2008). A two-component fit to the bright phase feature gave equivalent widths of −2.21 ± 0.02 Å for the emission and 1.47 ± 0.01 Å for the absorption, while a single component fit gave an equivalent width of −2.10 ± 0.02 Å for emission only. However, the high uncertainty in the corresponding line fluxes makes it difficult to determine whether the absorption is significant.

The He i emission profile has its highest total flux in the intermediate spectrum and its lowest in the December faint phase spectrum, indicating variable emission with orbital phase. Furthermore, the line strengths between the two faint spectra are not consistent with each other. It is possible that a heating event is responsible for the increased emission in the intermediate spectrum and/or the November faint spectrum. Zeng et al. (2014) studied enhanced He emission at ultraviolet (UV) and near-IR wavelengths following a C-class solar flare and attributed the increase in line flux to photoionization followed by recombination during the flaring event. Young stellar objects are highly variable objects that can experience frequent flaring events, which would increase the strength of He i emission for a brief time afterwards.

It is difficult to determine with certainty whether the varying He i emission strengths in our data can be attributed to flares, but any heating events that occur could be expected to cause an increase in the strengths of other features as well. The hydrogen Paschen α and Brackett γ features are observed in the bright and intermediate phase spectra, implying that they should at least be detectable in the November faint spectrum. However, only one of the two lines that were measured in the bright and intermediate spectra is distinguishable in the November spectrum, and neither line was present in the data from December. The heating event that caused the strengthening of the He i feature may have had either no impact or an inverse impact on the atomic hydrogen in the stellar magnetosphere. Perhaps the temperature increase implied by the He i outburst was sufficient to increase the degree of ionization of the magnetospheric hydrogen and actually reduce the amount of emission in the H_I features. We leave further examination of this interesting possible anti-correlation to future work.

The prominent H₂ ro-vibrational emission features appearing in the intermediate and faint phase spectra have previously been attributed to the bipolar outflow associated with the system's jet (Deming et al. 2004; Tokunaga et al. 2004). Table 7 lists the 10 H₂ lines that were identified in the intermediate and faint phase spectra, with the five features measured by Deming et al. (2004) denoted by asterisks. Equivalent widths were measured for all 10 lines, although values for several lines that were noisy in the intermediate and December faint phase spectra have been omitted. The continuum emission, F_c, was assumed to be flat in the immediate vicinity of each line and was estimated by fitting a constant to the region.

Figure 7 shows that the contrast between the H₂ lines and the continuum appears to increase as KH 15D becomes fainter, in accordance with expectation if the emission lines are arising from a source well away from the star itself. Although the November and December faint spectra were obtained at similar phases, the December lines consistently have smaller equivalent widths (see Table 8). The wavelengths of the H₂ emission features are contained within the K band, so broadband photometry from the two nights that the spectra were collected could help to account for the deviation in line strengths between cycles. However, the K band photometry from the December observing night was highly uncertain and cannot be used as a reliable measurement.

Zoom In Zoom Out Reset image size

The absolute fluxes for each H₂ line in the three spectra are presented in Table 8 and show that, on average, the line fluxes are more consistent between the intermediate and December spectra than between the November and December spectra. As discussed previously, cloudy weather during the November observing run resulted in a poor flux calibration for that spectrum. It is also possible that poor seeing conditions induced by the cloudiness may have scattered more light from the extended jet region into the slit in November than in the December or intermediate spectrum. This could lead to an apparent increased flux in the line. In any event, the uncertainties introduced by the flux calibration or variable extent of the region probed are minimized when the line flux ratios of the H₂ features are analyzed, as is the case here.

Deming et al. (2004) attributed the H₂ features to shock heating of ambient hydrogen within a thermally excited bipolar outflow, based on the measured line intensity ratio 2 − 1 S(1)/1 − 0 S(1) = 0.24 ± 0.05. This result is well below the value of ∼0.5 expected for fluorescent lines. Deming et al. (2004) measured the intensity of five of the hydrogen lines in a spectrum acquired at phase ∼0.17 (similar to our intermediate phase) and found a best-fit temperature of $T\sim 2800\pm 300$ K for the emitting region. However, it was noted that the observed line strength ratios did not agree with the expected values for this temperature within the measured uncertainties.

Table 9 gives the line flux ratios of the H₂ features in the most recent spectra (calculated from the absolute line fluxes in Table 8). The intensity of the 1−0 (Q1) line $(\lambda =2.4066\,\mu {\rm{m}})$ was used as the normalizing flux instead of the 1−0 (S1) line $(\lambda =2.1218\,\mu {\rm{m}})$, which had a low S/N in the December spectrum. Although the equivalent width of the 1−0 (Q3) line was greater than that of the 1−0 (Q1) line in all three spectra, the uncertainties were lower for 1−0 (Q1). The 2−1 (S1) feature $(\lambda =2.477\,\mu {\rm{m}})$ could only be measured for the November spectrum. We obtained a ratio of 2−1 (S1)/1−0 (S1) = 0.10 ± 0.02, which again confirms that the source of the emission is primarily shock excitation.

Deming et al. (2004) compared the spectrum obtained at phase ∼0.17 to a spectrum taken out of eclipse (similar to our bright phase) in which the 1−0 (S1) feature was resolved. The strength of the feature in the bright spectrum was much weaker than it was when the star was fainter, and the difference in strength was attributed to the increase in continuum flux when the stellar surface was revealed. If this were true in our spectra, we would expect the line flux ratios to remain roughly constant with orbital phase, despite the difference in absolute flux. The values in Table 9 for each line do not always agree within the measured uncertainties for all three spectra. However, we do not know the exact location of the emitting region detected in each of the three spectra and could simply be probing slightly different areas.

A thermal excitation temperature $(T)$ can be determined by assuming $\mathrm{ln}\left(\tfrac{F}{{gA}}\right)=-E/T$, with the properties of the 1−0 (Q1) line used for reference. Table 7 lists the line parameters A, g, and E for each of the 10 H₂ lines we observed. The most reliable flux ratios from Table 9 were used to determine the temperature of the emitting region. The resulting temperatures for the shock-excited H₂ are 2400 ± 300 K (November), 3000 ± 200 K (December), and 2800 ± 500 K (intermediate). The value obtained from the December spectrum is likely more uncertain than ±200 K, given the low S/N of some of the features. The temperatures from the November and intermediate spectra are consistent with each other and the 2800 ± 300 K result given by Deming et al. (2004). At the time the Deming et al. (2004) observations were conducted, the entire surface of star A was directly visible near apastron, and star B was fully occulted at all phases. When our spectra were acquired in 2013, star A was fully occulted at all phases and only part of star B was directly visible near apastron. The geometry of the system was quite different when the two sets of spectra were obtained, so the consistency between temperatures measured during two different epochs further confirms that the emission comes from the base of the bipolar outflow.

Figure 8 compares the measured flux ratios to the expected values for H₂ at the best-fit temperatures. Lines with $\tfrac{{\sigma }_{{F}_{\mathrm{line}}/{F}_{\mathrm{ref}}}}{{F}_{\mathrm{line}}/{F}_{\mathrm{ref}}}\geqslant 0.20$ have been omitted. The December spectrum only had two features that fell under this limit, again implying that the uncertainty in the temperature obtained from these data is actually greater than for the other two spectra. Although the measured flux ratios do not agree with the best-fit temperature for each individual feature, there is strong agreement between the values we have obtained from the spectra and what is expected for shock-excited H₂ in a bipolar outflow.

Zoom In Zoom Out Reset image size

To summarize this discussion, we find no convincing evidence for variations in the flux of the H₂ lines either on the orbital period or over the longer timescale represented by the difference in epochs between these data and the original studies. The current observations were not designed to study the jet emission, which is quite extended and should be investigated with a long slit aligned along the jet. Here the exact region probed by the H₂ lines is less clear, but it is apparent that it does not suffer variable obscuration by the ring. The line ratio data support the previous interpretation that it is shock heated ambient gas.

The scattered light from KH 15D reaches a maximum during the intermediate phase (Agol et al. 2004; Silvia & Agol 2008; Arulanantham et al. 2016), when direct starlight has declined but the bulk of the optically thick ring material is not yet between us and the star. To check whether the scattered light carries spectral signatures of the dust, reflectance spectra were created from the GNIRS data based on the procedure of Vilas et al. (1984), who removed the solar contribution from spectra of the surface of Mercury by dividing out a standard spectrum of the Sun. Debes et al. (2013) applied a similar procedure to near-IR spectra of TW Hya. Analogously, the intermediate phase spectrum of KH 15D was divided by the bright phase spectrum, where the features of star B are dominant. The most prominent emission features, such as He i from the magnetosphere and H₂ from the inner jet, were removed from the spectrum, which was normalized to the flux level at 0.8 μm. A 50 point moving average was applied to the data to increase the S/N of the remaining broad continuum while preserving the resolution of the original spectrum.

The slope of the resulting reflectance spectrum decreases toward longer wavelengths (see Figure 9, top panel), which is consistent with the forward scattering profile that is expected when the star is at the edge of the ring but not yet completely occulted (Arulanantham et al. 2016). Forward scattering at shorter wavelengths peaks at a higher relative flux level than the contribution from longer wavelengths near the ring edge, which causes the system to become bluer at this phase. The spectral slope is also indicative of reflectance by water ice. It is not as steep as the spectra of many icy bodies in the solar system, which are typically darkened and reddened by the presence of irradiated and/or exogenic organic, non-icy compounds (Cruikshank et al. 2005 and references therein). The spectrum is more consistent with small ice grains. Filacchione et al. (2012) found that the slope of modeled reflectance spectra of pure water ice becomes shallower with decreasing grain size. The slope of the KH 15D spectrum appears to be more similar to that of the near-IR spectra of Comet 17P/Holmes, whose features were attributed to a population of micron-sized grains in the coma (Yang et al. 2009).

Zoom In Zoom Out Reset image size

To further test the hypothesis that small water ice grains are a primary component of the reflecting material in the disk, and to constrain grain sizes, we have performed Mie scattering calculations for spherical ice grains with a variety of sizes. We used the mie_single code (http://eodg.atm.ox.ac.uk/MIE/) with optical constants for ice from Warren (1984). In Figure 10, we show the scattering efficiency of particles of different sizes over the relevant wavelength range. Grains of ∼1 μm in size scatter very efficiently in the blue end of the range, while larger particles have a flatter spectral response and begin to show the characteristic absorption bands at 1.5 and 2.0 μm. We require a mixture of particle sizes, including some as small as 1 μm and some much larger, to account for the observed blue slope and absorption features.

Zoom In Zoom Out Reset image size

After some exploration, we found that a particle size distribution that works well is a log-normal distribution with a peak at 1 μm and a σ of 2.0. The bulk scattering efficiency of spherical ice grains with that size distribution is shown in Figure 11. It reproduces the observed blue slope and 1.5 and 2.0 μm absorption features rather well. We note that most of the mass in such a distribution would be in large grains >20 μm, although there are many more small grains. A more detailed modeling effort is postponed until we can obtain a larger spectral range for the reflectance spectrum.

Zoom In Zoom Out Reset image size

The KH 15D spectrum also shows shallow absorptions near 1.04, 1.25, 1.55, and 2.02 μm (see Table 10 and Figure 9, bottom panel), which are the locations of prominent near-IR features of water ice (e.g., Clark & McCord 1980). All four absorptions have been measured in spectra of the larger Jovian and Saturnian satellites and Saturn's rings (Clark & McCord 1980; Cruikshank et al. 2005; Filacchione et al. 2012). The exact positions, shape, and depth of these bands is a function of grain size, crystallinity, and impurities. We conclude from this exercise that both the slope and primary absorption features of the observed reflectance spectrum can be understood as scattering from water ice grains with a distribution of grain sizes from about 1–100 μm. In order to confirm the detection of water ice in KH 15D, observations should be extended out to ∼5 μm. This will enable measurements of the 3 μm ice absorption that was used to set a lower limit on the abundance of ice in Comet 17P/Holmes (Yang et al. 2009).

In addition to the water ice absorptions, the KH 15D spectrum has spectral features that are characteristic of hydrocarbons (Cloutis 1989), specifically methane ice (Quirico & Schmitt 1997; Grundy et al. 2002; Clark et al. 2009). Sublimation temperatures for these two constituents are around 160 and 45 K (Dodson-Robinson et al. 2009), implying that the temperature in the scattering zone is very cold. Based on its precession rate, the KH 15D circumbinary ring is centered at a distance of 3.9 au from the binary center of mass (Arulanantham et al. 2016). Its full width is expected to be approximately the same as its radial distance, although this dimension is much less certain than the radial distance. We may infer that the ring stretches roughly from 2 to 6 au from the central stars, whose combined luminosity are currently $\sim 1\,{L}_{\odot }$. It is unknown whether the occulting edge of the ring is its inner edge, its outer edge, or somewhere in between (Winn et al. 2006). It is therefore plausible that the scattering zone of the circumbinary ring would be outside the water frost line, but not the methane frost line, unless the grains are shielded from the direct radiation of the central stars.

Dodson-Robinson et al. (2009) have explored models of ice line evolution in young disks that may be applicable to our observations. They find that by the age of the KH 15D system (∼2 Myr) the methane ice line has moved in to about 5 au, while the water ice line is at about 1 au. Their model predicts that near 6 au, coincident with the inferred position of the outer edge of the KH 15D circumbinary ring, the two most abundant ice species, by far, would be water and methane, which is just what we observe. It is uncertain how much relevance their model has, since it is for a fairly massive disk around a single star of 0.95 M_⊙, which is less than the combined mass of the KH 15D binary $(1.3\,{M}_{\odot })$. It also does not include the effect of settling by large dust grains, which may increase the shielding, providing a colder environment even closer to the central stars.

To summarize, it is not surprising to find water ice as a principal component of the scattering grains in KH 15D, given the likely distance of the scattering zone from the stars, but it is somewhat surprising to find methane ice. While it could form and survive at 5 au or even closer if shielded from the central starlight, it must be exposed to that starlight in order to actually scatter it. One would expect rapid sublimation of such ice exposed to the stellar radiation. As noted above, the required log-normal distribution of grain sizes suggests that most of the mass of the ice is in large grains, even pebbles or boulders. More sophisticated models of disk evolution may be required to account for our observation that water and methane ice are the dominant species in the KH 15D circumbinary ring.

The GNIRS data were obtained at various orbital phases with the goal of isolating spectral signatures of the putative planet from features of the stars. The best opportunity to accomplish this comes during the faint phase, when the contribution to the continuum spectrum from scattered starlight is at a minimum. An average continuum value was determined in three different regions of the faint and intermediate phase spectra (corresponding to the J, H, and K bands), and scale factors were determined for each region by dividing the continuum values from the faint spectra by those from the intermediate spectrum. The intermediate spectrum was then multiplied by the appropriate scale factors and subtracted from each of the two faint phase spectra in order to remove any residual features from the stars and the ring.

Figures 12 and 13 illustrate the accuracy of this method by zooming in on three different regions of the faint phase spectra and comparing them to the scaled intermediate spectrum in the same wavelength ranges. A 50 point moving average was applied to the star-subtracted faint phase spectra in order to increase the S/N after the prominent emission lines were removed. The resulting faint phase "planet" spectra are plotted in Figure 14.

Zoom In Zoom Out Reset image size

Zoom In Zoom Out Reset image size

Zoom In Zoom Out Reset image size

The scale factors that were applied to the intermediate spectrum have, for the most part, reduced its flux level to those of the faint spectra, although there is a small over-correction at shorter wavelengths that results in negative flux. This procedure could be improved by using a detailed model to predict what the scattered light spectrum will look like during the faint phase. It is also possible that the negative flux is due to variable telluric absorption, which is difficult to correct for in low S/N data. The telluric features are expected to be present at 0.89–0.96 μm, 1.11–1.17 μm, 1.25–1.27 μm, and 1.32–1.5 μm. These regions have been plotted in red in Figure 14, and we caution the reader against over-interpreting the planet spectra at these wavelengths.

Each faint spectrum was compared to model spectra of young giant planets from Spiegel & Burrows (2012) to see if any continuum features were similar to what is expected for a 1 Myr, 10 M_J planet (see Figure 14). The December spectrum in particular has a significant peak in emission near ∼2 μm and smaller bumps at the locations of other peaks in the model spectra. The features at 1.6 and 2.1 μm agree with model spectra of NH₃ and/or CH₄ (Sharp & Burrows 2007), while the features near 1.0 and 1.3 μm, if they are real, could be due either to H₂O or CH₄. As described in Section 3.4, detailed models of scattering and absorption signatures from the minerals we have detected in the circumbinary ring will require extending our observations out to ∼5 μm. Such models will be crucial to ensuring that the putative planet spectra can be accurately separated from the other components in the system.

It is also possible that the increased reddening observed near minimum light is due to emission from dust in the ring, rather than a giant planet. However, Windemuth & Herbst (2014) determined that the amount of excess H band flux was consistent with a ∼1000–2000 K source. Our analysis of the scattering properties of the circumbinary ring indicates that the dust temperature is much colder than even 1000 K, but longer wavelength observations are required to constrain this.