Evidence for Abnormal Hα Variability During Near-transit Observations of HD 189733 b

Much effort has been invested into understanding how stellar activity affects radial velocity measurements and broadband transit observations (e.g., Saar & Donahue 1997; Pont et al. 2011; Aigrain et al. 2012; Dumusque et al. 2014; Oshagh et al. 2014; Andersen & Korhonen 2015; Llama & Shkolnik 2015; Hébrard et al. 2016; Herrero et al. 2016; Chiavassa et al. 2017). There are many studies on long-term activity variations, i.e., day to year timescales, for known exoplanet hosts (Boisse et al. 2009; Fares et al. 2010; Gomes da Silva et al. 2011, 2014; Figueira et al. 2016; Giguere et al. 2016). On the other hand, little investigation has been aimed at understanding stellar activity variations on very short timescales for known exoplanet host stars (i.e., minutes to hours). This is unsurprising, as long-term variations play a more significant role in RV data sets collected sporadically across many nights. Similarly, even large variations in spectroscopic activity indicators (e.g., Hα, Ca ii, or Na i) are completely washed out in broadband photometric observations. With the increasing popularity of short-cadence, high-spectral resolution transit observations (e.g., Redfield et al. 2008; Jensen et al. 2012; Cauley et al. 2015, 2016; Wyttenbach et al. 2015; Barnes et al. 2016), it is important to understand the short-term behavior of host stars in order to differentiate between true absorption by planetary material and stochastic changes in stellar activity indicators potentially caused by star–planet interactions (SPIs).

In Cauley et al. (2015, 2016), we reported on pre- and post-transit Hα absorption signatures in high-resolution data taken across two different transits of HD 189733 b. Similar signatures have been reported in atomic UV transitions for HD 189733 b (Ben-Jaffel & Ballester 2013; Bourrier et al. 2013) and WASP-12 b (Fossati et al. 2010; Haswell et al. 2012). Each of the Hα signatures showed a different depth and time series shape, suggesting that the physical mechanism is highly variable. Because absorption measurements from high-resolution transmission spectra are necessarily normalized relative to a reference spectrum, or a group of spectra, the absorption is, by definition, relative to another point in time. Thus, if the reference spectrum is chosen during a time of higher stellar activity, other spectra will show absorption, mimicking the loss of photons by line of sight material. The short baselines, i.e., a single night, for the high-cadence transit observations make it difficult to distinguish between these changes in the stellar activity level and true absorption by circumplanetary material.

In this paper, we present out-of-transit observations of HD 189733 in order to further probe the frequency of the observed pre- and post-transit changes in the Hα line core. Section 2 describes the observations and data reduction procedures. In Section 3, we present the Hα time series data and discuss the statistical analysis of the changes in the Hα signal. A brief summary is given in Section 4.

Observations were performed using the Tull Coudé Spectrograph (Tull et al. 1995) on the Harlan J. Smith 2.7 m telescope at McDonald Observatory. We collected data on five separate nights. A to-scale diagram of the phases during which the observations took place is shown in Figure 1. The night of 2016 July 29 was shortened due to persistent high humidity, and 2016 September 19 was shortened because of heavy cloud cover during the first two hours of the night. The nights of 2016 July 31 and 2016 August 01 were shortened due to observatory functions requiring use of the telescope during the first third of each night. Details concerning the 2013 and 2015 Keck observations are presented in Cauley et al. (2015, 2016).

Zoom In Zoom Out Reset image size

Spectra were taken with two different slits. We utilized the #5 slit for the first night of 2016 July 29, which has a width of 179 and resolving power of $R\sim {\rm{40,000}}$, or ∼7.5 km s⁻¹. These observations were typically 900 s in length. For the nights of 2016 July 30 and July 31, we used the #6 slit with $R\sim {\rm{30,000}}$ (or ∼10 km s⁻¹) and 600 s integrations in order to use a more similar cadence to the past Keck observations and to boost the signal-to-noise. Exposure times were increased to 900 s on 2016 August 01, due to moderate cirrus cloud coverage. The telluric standard HR 8634 was observed for approximately ten minutes each night. The spectra from 2016 July 29 were broadened to match the instrumental resolution of the observations taken with the #6 slit in order to standardize the data collected with different spectrograph settings.

Standard data reduction steps were taken, including bias subtraction, flat fielding with a median flat, and optimal extraction using custom IDL routines. Spectra were extracted for two separate orders, one containing the Hα line and the other containing a comparison Fe i line at 6677.98 Å. The Fe i line is used as a control to ensure that the extraction and spectrum comparison procedures are not artificially producing signals in the line cores. Wavelength solutions for the individual orders were found using Th–Ar lamp exposures and 3rd or 4th degree polynomial fits. Typically ∼20–30 Th–Ar lines were used in the solution. Small wavelength shifts between individual spectra are corrected using strong stellar lines, besides the line of interest, in the same order. Each wavelength solution is then corrected for the barycentric velocity and HD 189733 system radial velocity, which we take to be −2.23 km s⁻¹ (di Gloria et al. 2015).

The telluric spectrum in the Hα order was modeled using the program Molecfit (Kausch et al. 2014). We modeled the telluric spectrum in the Fe i order, but it is very weak and results in negligible changes to the transmission spectrum. We first removed the blaze function and the broad stellar Hα line from the telluric standard spectrum using a high order spline. Molecfit then fits the normalized telluric spectrum for the H₂O and O₂ column densities, as well as the instrumental resolution. This master telluric model is then scaled, shifted, and divided out of the individual observations.

The mean normalized Hα spectrum from each night is shown in the top row of Figure 2. We also show the ratio of the master spectrum from each night compared with the master spectrum from 2016 September 19 in the middle row (see Equation (1)). The middle rightmost panel shows the 2016 September 19 spectrum compared with the 2013 and 2015 master spectra. The equivalent width measurement ${W}_{{\rm{H}}\alpha }$ (see Equation (2)) for the ${F}_{i}/{F}_{\mathrm{Sep}19}-1$ spectra are given in the bottom left of the middle panels, where negative ${W}_{{\rm{H}}\alpha }$ indicates a deeper Hα core relative to 2016 September 19, i.e., less core emission. The night-to-night ${W}_{{\rm{H}}\alpha }$ values can be compared to the intra-night changes shown in Figures 4 and 7. We also show histograms of the transmission spectra for 50 km s⁻¹ $\lt | v| \lt 200$ km s⁻¹ in the bottom panels. The same comparison for the Fe i control line is shown in Figure 3.

Zoom In Zoom Out Reset image size

Zoom In Zoom Out Reset image size

Zoom In Zoom Out Reset image size

Figure 2 shows that all of the July nights exhibit less Hα core emission than the 2016 September 19 observations, suggesting that HD 189733 was in a more active state, or the visible hemisphere was more active, during the night of 2016 September 19. There is some fluctuation between the July nights, with 2016 July 30 showing the most core emission and 2016 July 29 and 2016 August 01 showing the least. The 2013 data shows a similar activity level to 2016 July 30, while the 2015 data shows a much more active state than any of the other dates, with ∼10% more Hα core emission when compared with 2016 August 01. In a forthcoming paper, we note that the 2015 observations are over-active compared to the other examined nights from that study. The additional data presented here strengthens the conclusion that HD 189733 was especially active at that time.

In order to produce a time series of the relative changes in the Hα core within an individual night, we produce a "transmission" spectrum that is identical to the definition of S_T from Cauley et al. (2015, 2016):

$$\begin{eqnarray}&&{S}_{T}=\displaystyle \frac{{F}_{i}}{{F}_{\mathrm{comp}}}-1\end{eqnarray} \tag{ 1 }$$

where F_i is a single observation and F_comp is the master comparison spectrum constructed from the three highest signal-to-noise spectra from each night. We then calculate the equivalent width of the individual transmission spectra according to

$$\begin{eqnarray}&&{W}_{{\rm{H}}\alpha }=\displaystyle \sum _{v=-200}^{+200}\left(1-\displaystyle \frac{{F}_{v}}{{F}_{v}^{\mathrm{comp}}}\right){\rm{\Delta }}{\lambda }_{v}\ \mathring{\rm A} ,\end{eqnarray} \tag{ 2 }$$

where F_v is the flux in the spectrum of interest at velocity v, F_v^comp is the flux in the comparison spectrum at velocity v, and ${\rm{\Delta }}{\lambda }_{v}$ is the wavelength difference at velocity v. Uncertainties in ${W}_{{\rm{H}}\alpha }$ are calculated by summing the transmission spectrum flux errors in quadrature across the same velocity range.

The phase-folded Hα time series for all nights is shown in Figure 4, and the Fe i time series is shown in Figure 5. A histogram comparison of ${W}_{{\rm{H}}\alpha }$ between the 2016 dates and the combination of the 2013 and 2015 dates is shown in Figure 6. The out-of-transit points from the 2013 and 2015 data sets are also shown in Figure 4. The magnitude of the pre-transit signals measured in Cauley et al. (2015, 2016) are marked with red and blue solid horizontal lines. It is clear that no changes in the out-of-transit monitoring data are of comparable strength to the 2013 pre-transit signal. Some changes, however, are of similar magnitude to the pre-transit absorption signals from 2015, presented in Cauley et al. (2016). The left panel of Figure 6 demonstrates that the 2016 ${W}_{{\rm{H}}\alpha }$ measurements are consistent with the random measurement errors and show no significant deviations from zero. The right panel of Figure 6 shows that there is no correlation between ${W}_{{\rm{H}}\alpha }$ and W_{Fe i}, suggesting that the extraction and transmission spectrum procedures are not significantly contributing to the measurements. The same ${W}_{{\rm{H}}\alpha }$ data is shown in Figure 7 but as a function of time from the midpoint of each set of observations.

Zoom In Zoom Out Reset image size

Zoom In Zoom Out Reset image size

Zoom In Zoom Out Reset image size

To quantify the differences in the measured ${W}_{{\rm{H}}\alpha }$ time series, we have calculated the absolute deviation from the median, or the ADM, for each night:

$$\begin{eqnarray}&&\mathrm{ADM}=| {W}_{{\rm{H}}\alpha }^{i}-\mathrm{Median}({W}_{{\rm{H}}\alpha })| \ \mathring{\rm A} \end{eqnarray} \tag{ 3 }$$

where ${W}_{{\rm{H}}\alpha }^{i}$ are the individual observations and Median (${W}_{{\rm{H}}\alpha }$) is the median of all observations. The empirical distribution function (EDF) of the ADM for each night is shown in the left panel of Figure 8. The right panel of Figure 8 shows the total EDFs for the 2016 data (green line) and the 2013 data combined with the 2015 data (magenta line). We also include normal and uniform distributions, generated using the 2016 data, for reference. We do not make any statistical comparisons using only the 2013 data, due to the smaller number of observations compared with the 2015 data set. The 2013 and 2015 EDFs are plotted with dashed red and blue lines, respectively, in the left panel. Also shown in the left panel of Figure 8 are the two-sided Kolmogorov–Smirnov (KS) statistic D_KS and the probability p_KS that the null hypothesis is true, i.e., that each EDF is drawn from the same parent distribution, for the individual 2016 EDFs compared with the 2015 EDF. The KS statistics are shown in the right panel for the comparison between both total EDFs.

Zoom In Zoom Out Reset image size

For our purposes, the EDF of the absolute deviation from the median highlights differences between a time series with structure, such as the 2015 data, and the apparently random deviations of a time series like those from the 2016 nights. Figure 8 shows significant differences between the 2015 EDF and the 2016 EDFs: the 2015 EDF flattens off near $\mathrm{ADM}\sim 0.0025$ and then begins to increase again near $\mathrm{ADM}\sim 0.0045$. In comparison, all of the 2016 EDFs are ∼80% comprised of $\mathrm{ADM}\lesssim 0.004$. These differences are born out of the KS statistics: with the exception of the 2016 July 29 observations, which contain only eleven data points, all of the EDFs differ from the 2015 EDF at the 95% level, i.e., ${P}_{\mathrm{KS}}\lt 0.05$. We can rule out the null hypothesis even more strongly ($\gt 99 \%$) for the combined EDF test. This can be interpreted as the 2016 dates all showing random deviations from the median activity level while the 2015 data has distinct, sustained features that force data points contained in these features to be far from the median. This contributes to the EDF at the higher ADM values.

The significant differences between the out-of-transit 2016 monitoring and the very near-transit data suggest that something interesting is happening immediately before and after HD 189733 b transits. There are two possibilities: (1) the near-transit features are due to absorbing circumplanetary material; or (2) the stellar activity level, as measured in Hα, is experiencing abnormal changes preferentially at near-transit times. Using the current data, it is not clear which of these two cases is true. However, we can say with some confidence that abnormal changes in the stellar activity level occur more frequently very close to planetary transits. In addition, the nominal stellar activity level is much higher for the 2015 data compared with the 2013 data set, and the 2013 data shows a similar activity level to the 2016 observations, yet both the 2013 and 2015 observations show significantly different ADM distributions compared with the 2016 data. Thus, we cannot attribute the abnormal near-transit ${W}_{{\rm{H}}\alpha }$ measurements to high levels of stellar activity. Taken together, we believe this is strong evidence in support of SPIs or absorption by circumplanetary material in the HD 189733 system (Cuntz et al. 2000; Shkolnik et al. 2008; Shkolnik 2013; Strugarek et al. 2014; Miller et al. 2015; Pillitteri et al. 2015).

Finally, we note that if the Hα variations are the result of SPIs near transit, the affected region on the star must be very near the sub-stellar point. We speculate that this is evidence for the circumplanetary absorption interpretation, as there is nothing special about the transit in the SPI interpretation. We leave this suggestion to further investigation.

We have presented new out-of-transit, high-cadence Hα monitoring of HD 189733 over the course of five nights using the Tull Coudé spectrograph on the 2.7 m Harlan J. Smith telescope at McDonald Observatory, with the goal of establishing the frequency of changes in the stellar activity level at a similar magnitude to what we presented in Cauley et al. (2015, 2016). We do not find any variations in the Hα core flux that are similar to what was observed immediately pre-transit in Cauley et al. (2015). With the exception of the 2016 July 30 data set, which only contains 11 observations, we find statistically significant differences between each of the out-of-transit nights and the out-of-transit signal from Cauley et al. (2016). This conclusion is strengthened when the combined out-of-transit data set is compared with the combined near-transit observations. Our results suggest that changes in Hα similar to those from Cauley et al. (2015) and Cauley et al. (2016) occur infrequently when the planet is far from transit. This is evidence for attributing the pre-transit signals to either absorbing circumplanetary material or some type of magnetic or tidal SPI near the sub-planetary point on the stellar surface. Further monitoring is necessary to understand the frequency and physical nature of the near-transit changes.

The authors thank the referee for their suggestions, which helped to significantly improve this manuscript. This paper includes data taken at The McDonald Observatory of The University of Texas at Austin. P.W.C. is grateful to Dave Doss and Brian Roman for their instrumental knowledge and observational support. This work was completed with support from the National Science Foundation through Astronomy and Astrophysics Research Grant AST-1313268 (PI: SR). A.G.J. is supported by NASA Exoplanet Research Program grant 14-XRP14 2-0090 to the University of Nebraska-Kearney. This work has made use of NASA's Astrophysics Data System.