A decade of H α transits for HD 189733 b: stellar activity versus absorption in the extended atmosphere
DDraft version March 29, 2017
Preprint typeset using L A TEX style AASTeX6 v. 1.0
A DECADE OF H α TRANSITS FOR HD 189733 B: STELLAR ACTIVITY VERSUS ABSORPTION IN THEEXTENDED ATMOSPHERE
P. Wilson Cauley and Seth Redfield
Wesleyan UniversityAstronomy Department, Van Vleck Observatory, 96 Foss Hill Drive, Middletown, CT 06459
Adam G. Jensen
University of Nebraska-KearneyDepartment of Physics & Astronomy, 24011 11th Avenue, Kearney, NE 68849
ABSTRACTHD 189733 b is one of the most well-studied exoplanets due to its large transit depth and host starbrightness. The focus on this object has produced a number of high-cadence transit observationsusing high-resolution optical spectrographs. Here we present an analysis of seven full H α transits ofHD 189733 b using HARPS on the 3.6 meter La Silla telescope and HIRES on Keck I, taken overthe course of nine years from 2006 to 2015. H α transmission signals are analyzed as a function of thestellar activity level, as measured using the normalized core flux of the Ca II H and K lines. We findstrong variations in the strength of the H α transmission spectrum from epoch to epoch. However,there is no clear trend between the Ca II core emission and the strength of the in-transit H α signal,although the transit showing the largest absorption value also occurs when the star is the most active.We present simulations of the in-transit contrast effect and find that the planet must consistentlytransit active latitudes with very strong facular and plage emission regions in order to reproduce theobserved line strengths. We also investigate the measured velocity centroids with models of planetaryrotation and show that the small line profile velocities could be due to large velocities in the upperatmosphere of the planet. Overall, we find it more likely that the measured H α signals arise in theextended planetary atmosphere, although a better understanding of active region emission for activestars such as HD 189733 are needed. INTRODUCTIONHot planets, or planets orbiting within ∼
10 stellar radii of their host stars and which have orbital periods of a fewdays, are unique subjects for planetary astrophysics. The extreme proximity to their host stars can result in phenomenathat are not observable in other exoplanet systems (for a recent review of these processes, see Matsakos et al. 2015).These phenomena include atmospheric mass loss (e.g., Vidal-Madjar et al. 2003; Murray-Clay et al. 2009; Ehrenreichet al. 2012; Bourrier et al. 2013; Ehrenreich et al. 2015; Khodachenko et al. 2015; Salz et al. 2016), magnetic and tidalstar-planet interactions (e.g., Cuntz et al. 2000; Shkolnik et al. 2008; Lanza 2009; Strugarek et al. 2014; Poppenhaeger& Wolk 2014; Pillitteri et al. 2015; Miller et al. 2015), and bow shocks that can form where the planet’s atmosphereor magnetosphere plows through the stellar wind (e.g., Lai et al. 2010; Vidotto et al. 2011; Ben-Jaffel & Ballester2013; Llama et al. 2011, 2013; Cauley et al. 2015; Turner et al. 2016). Due to the larger magnitude of the potentiallyobservable effect, giant hot planets ( M (cid:38) M Neptune ) are particularly good targets for investigating these processes.One of the most well studied hot Jupiters is HD 189733 b (Bouchy et al. 2005). Due to the brightness of its host star( V = 7 .
7) and its large transit depth ( ∼ α and concluded that the absorbing material must be gravitationally unbound. A followup study showed [email protected] a r X i v : . [ a s t r o - ph . E P ] M a r that the evaporation is highly variable: no absorption was detected in a set of 2010 Lyman- α observations while astrong Lyman- α transit, with absorption up to ∼
14% measured in the blue wing of the line profile, was detected in2011 (Lecavelier des Etangs et al. 2012). These results were the first indication of variability in the mass loss of a hotexoplanet.Evaporation of hot planets is driven by atmospheric heating from the absorption of X-ray and extreme UV (EUV)stellar radiation (e.g., Murray-Clay et al. 2009; Owen & Jackson 2012). Two mechanisms can thus produce variationsin the amount of X-ray and EUV flux received by a hot planet: 1. stellar rotation and the planet’s orbital motion,which cause active regions of differing strengths to be directed toward the planet throughout its orbit; and 2. intrinsictime variability in the stellar activity level due to long-term activity cycles or short term variability, such as flares.These variations provide a natural explanation for changes in the planetary evaporation rate and suggest that hotplanets subjected to larger amounts of ionizing stellar radiation will have higher evaporation rates (Owen & Adams2016).Measuring the exosphere of hot planets requires space-based UV observations (e.g., Fossati et al. 2010; Ben-Jaffel& Ballester 2013; Bourrier et al. 2013). However, the extended atmosphere, or the thermosphere at pressures of 10 − - 10 − bar, can be observed from the ground using the neutral hydrogen Balmer line transitions (Jensen et al. 2012;Christie et al. 2013; Astudillo-Defru & Rojo 2013; Cauley et al. 2015, 2016). The first H α detection was made byJensen et al. (2012). Christie et al. (2013) modeled the Jensen et al. (2012) detection and showed that the strengthof H α absorption in the atmosphere of a hot Jupiter is dependent on the amount of ionizing stellar radiation, withlarger amounts of radiation producing stronger absorption. Thus while H α observations most likely do not directlyprobe the evaporating material, the strength of absorption in the extended atmosphere may provide insight into thestrength of the evaporation.Recently, Barnes et al. (2016) used archival HARPS data to measure in-transit H α transmission spectra for HD189733 b across three separate transits. Based on velocity maps of the absorption signatures in both the stellar andplanetary rest frames, the authors point out that the absorption signal moves from red-shifted to blue-shifted velocitiesin the planetary rest frame and shows no velocity gradient across the transit in the stellar rest frame. They suggestthat this is evidence of the absorption signature arising in the frame of the star, i.e., it is not absorption by planetarymaterial but rather the result of continuum-to-line contrast effects that arise as the planet occults different portions ofthe stellar disk. They also highlight the potential problems with using continuum-to-line differential measurements forchromospherically sensitive lines: a non-uniform stellar surface with active regions can be weighted towards a strongeror weaker line core based on which portion of the stellar surface is occulted by the planet, resulting in a contrast effectbetween the line core and nearby continuum region (Berta et al. 2011). Barnes et al. (2016) conclude that the transitsignatures measured in the chromospherically sensitive lines, in particular H α and the Ca II H and K lines, are theresult of the contrast effect and not due to absorption by planetary material. This is especially relevant to active starssuch as HD 189733 ( S HK = 0 .
52; compared to the solar value of S HK = 0 .
17; Wright et al. 2004).In this paper we examine seven archival transits of HD 189733 b, including the same HARPS data analyzed byBarnes et al. (2016), in order to search for a relationship between the stellar activity level and the strength of thein-transit H α signal. We do not include the results of Jensen et al. (2012) since these observations were not performedacross a single transit. We produce detailed transit models in order to investigate the contrast effect and determine ifthe H α signatures measured in Jensen et al. (2012), Cauley et al. (2015), Cauley et al. (2016), and Barnes et al. (2016)can be attributed to occultation of a non-uniform stellar surface rather than absorption by planetary material. We alsodiscuss how planetary rotation can affect the measured absorption velocities, which we calculate for each transit, andpresent transmission spectrum models of rotation in an extended planetary atmosphere. The investigation into theabsorption line velocities is motivated by the suggestion of Barnes et al. (2016) that the H α transmission is not due tothe planetary atmosphere. Their main argument is that the H α line velocities do not follow the pattern expected fromabsorption in the planetary atmosphere, although we note that they do not present quantitative measurements of theline velocities. The data sets and reduction processes are described in Section 2. The transmission spectrum is definedin Section 3 and the average H α transmission spectra are given. H α and Ca II H and K time-series measurements arepresented in Section 4 for each transit. Section 5 includes an examination of epoch to epoch changes in H α absorptionand the stellar activity level as measured using the Ca II H and K lines. The contrast models are discussed in Section 6and the atmospheric rotation models and measured line velocities are given in Section 7. Section 8 provides a briefsummary and conclusion of our results. Throughout the paper we refer to active regions as portions of the stellar surface covered by spots or faculae and plage. OBSERVATIONS AND DATA REDUCTIONThe observations presented here are a combination of HD 189733 b transits observed with HARPS (Mayor et al.2003) on the 3.6-meter telescope at La Silla and HIRES (Vogt et al. 1994) on Keck I. Information about each data setis detailed in Table 1, including the average signal-to-noise (S/N) of an individual exposure in the continuum near 6571˚A. All data is currently available on either the ESO data archive or the Keck Observatory Archive. The archive ID foreach data set is given in Table 1. We note that the HARPS data is the same used by Wyttenbach et al. (2015) andLouden & Wheatley (2015) to study Na I absorption in HD 189733 b’s atmosphere. The Keck data from 2006 August21 is the same data used to measure the Rossiter-McLaughlin effect for HD 189733 b by Winn et al. (2006). We alsoretrieved an archival transit obtained using UVES (Dekker et al. 2000) by Czesla et al. (2015) to study center-to-limbvariations of the Na I D lines. This transit, however, is fundamentally different when compared to all of the other H α transits and does not match any of the features seen in the HIRES and HARPS time series. This abnormal behavior isdue to the mid-transit flare identified by Czesla et al. (2015) and further investigated by Klocov´a et al. (2017). Finally,we exclude archival transits from the High Dispersion Spectrograph on Subaru due to the lack of simultaneous Ca II observations.HARPS has a resolving power of R ∼ ,
000 and the HIRES observations were performed at R ∼ ,
000 usingthe B5 decker and R ∼ ,
000 using the B2 decker. The reduced HARPS data were taken directly from the ESOarchive. We note that we do not perform the scattered light removal process described by Barnes et al. (2016). Webelieve that a comparison of our timeseries results and average transmission spectra provide justification: we obtainalmost identical results for the behavior of both H α and Ca II using the standard HARPS reduction routines.Standard reduction steps including bias subtraction, flat fielding, and wavelength calibration were performed for theKeck data using the publicly available HIRES Redux program by Jason X. Prochaska . All spectra are shifted to therest frame of the star by correcting for the Earth’s barycentric velocity and HD 189733’s system radial velocity, forwhich we use the mean value − .
23 km s − from Di Gloria et al. (2015).We used the latest version of Molecfit (Kausch et al. 2014) in order to model telluric absorption in the H α order. Wefirst construct a master telluric model by using a telluric standard observed on the same night. The master telluricmodel is then fit to a selection of telluric lines in the individual science exposures using a χ minimization routine.We fit for the line depth, small wavelength shifts, and line broadening. The best-fit scaled, broadened, and shiftedtelluric model is then divided out of the normalized science spectrum. This routine works very well for most spectra,removing the telluric absorption down to 5 - 10% of the original line depth. This typically results in transmissionspectrum residuals of ∼ α for the 2007 August 29 observations (seeFigure 1). This portion of the spectrum is not included in the absorption calculations.Two-element wavelength binning is applied to all of the individual HARPS spectra in order to increase the signal-to-noise (S/N). We also co-add back-to-back spectra from the nights of 2007 July 20 and 2007 August 29 due tothe short exposure times (300 seconds). Note that the S/N values for the HARPS spectra in Table 1 are calculatedfor the co-added and binned spectra. The individual Keck observations required no binning. Some spectra near thebeginning or end of the night are not included in the analysis due to S/N (cid:46)
50 or contamination by Earth’s twilightsky spectrum.
Table 1 . Archived data sets
UT Date Instrument Program ID N obs N used S/N @ 6571 ˚A R λ start , λ end (1) (2) (3) (4) (5) (6) (7) (8)2006 Aug 21 Keck HIRES A259Hr 70 55 230 49,000 3360˚A,7800˚A2006 Sep 8 HARPS 072.C-0488(E) 18 18 149 115,000 3800˚A,6800˚A2007 Jul 20 HARPS 079.C-0828(A) 39 39 140 115,000 3800˚A,6800˚A2007 Aug 29 HARPS 079.C-0127(A) 40 40 131 115,000 3800˚A,6800˚A2013 Jun 3 Keck HIRES A308Hr 17 16 440 68,000 3500˚A,7400˚A2013 Jul 4 Keck HIRES A308Hr 40 37 490 68,000 3500˚A,7400˚A2015 Aug 4 Keck HIRES N120Hr 61 61 490 68,000 3500˚A,7400˚A ∼ xavier/HIRedux/ TRANSMISSION SPECTRAThe transmission spectrum is defined here as: S T = F i F out − F i is a single in-transit observation and F out is the master comparison spectrum. To produce the final transmis-sion spectra, we apply the same wavelength alignment and normalization procedures described in Cauley et al. (2015,2016). The comparison spectra are generally chosen to be those taken furthest from the transit, although there is nostrict rule for doing so, and their number is selected to minimize the impact of any single spectrum while not using toomany of the available observations as comparison exposures. For example, the 2006 August 21 comparison spectra arechosen as those immediately before the transit in order to balance the low W Hα values near t − t mid = −
100 minutesand the increase in W Hα immediately after the transit begins. The choice of comparison spectra can significantlyinfluence the absolute level of measured absorption but the relative changes between observations remain the same.For this study, the choice of out-of-transit comparison spectra does not significantly change the results for any of thetransits except the 2015 Aug 4 data. The choice of comparison spectra for 2015 Aug 4 is detailed in Cauley et al.(2016).The average H α in-transit transmission spectrum for the individual transits is shown in the right side column ofFigure 1. Only individual in-transit observations showing 1 σ significant absorption are selected to be included in theaverage transmission spectra in order to highlight the line morphology. Including all of the in-transit observations from2013 June 3, for example, results in a much weaker transmission spectrum due to the abrupt decrease in absorptionmid-transit. Master absorption measurements (see Equation 2) are given for each date in Table 2. Absorption isdetected at the 3 σ level for all transits.Significant variations in S T from epoch to epoch are evident, especially in the high S/N Keck spectra. The HARPSspectra are fairly noisy but clear changes in line depth, and even shape, can be seen. These variations suggest that ifthe signal arises in HD 189733 b’s extended planetary atmosphere, it changes rather drastically from one epoch to thenext, and perhaps within individual epochs, as observed in the Ly α exosphere by Lecavelier des Etangs et al. (2010)and Lecavelier des Etangs et al. (2012) and across the multiple H α transits observed by Jensen et al. (2012). TIME-SERIES H α ABSORPTION AND THE NORMALIZED CA II CORE FLUXFor each individual spectrum we calculate the following absorption measure, essentially an equivalent width of thetransmission spectrum, at H α : W Hα = +200 (cid:88) v = − (cid:18) − F v F outv (cid:19) ∆ λ v (2)where F v is the flux in the spectrum of interest at velocity v , F outv is the flux in the comparison spectrum at velocity v , and ∆ λ v is the wavelength difference at velocity v . The units of W Hα are angstroms. The gray shaded region inthe transmission spectrum for 2007 August 29 is ignored due to poor telluric subtraction.Individual HARPS transmission spectra are normalized across the ±
200 km s − region by averaging the fits of aline and a low order spline. The ±
40 km s − at line center are ignored in the normalization. Individual Keck spectra,which are much higher S/N than the HARPS transmission spectra and have fewer telluric residuals, require only asecond-order polynomial to adequately remove the continuum slope. We note that small residual offsets from zeroin the normalized HARPS transmission spectra can result in W Hα offsets of ∼ W Hα values calculated for the individual comparison points. For this reason, we do not separatelyinclude the normalization uncertainties.We derive uncertainties in W Hα by combining in quadrature two different sources of uncertainty. First, normalizedflux errors in the transmission spectrum are summed in quadrature. This is then added in quadrature to the standarddeviation of the comparison spectra W Hα points (purple circles in Figure 1). The standard deviation uncertaintiesdominate the normalized flux uncertainties in most cases. We note that this is different from the empirical Monte Carlo(EMC) procedure used in Redfield et al. (2008), Jensen et al. (2012), Cauley et al. (2015), and Cauley et al. (2016)but has a similar outcome: large variations in the comparison spectra, like in the 2013 July 4 transit, will producelarger uncertainties in all of the individual points. We choose to use the standard deviation of the comparison pointssince the HARPS time series, compared with the 2013 and 2015 Keck time series, have relatively fewer comparisonspectra ( N = 4 − N = 8 for the 2013 and 2015 Keck nights). The EMC procedure becomes less useful −300 −200 −100 0 100 200t−t mid (minutes)−0.0150−0.00750.00000.00750.0150 − W H α ( Å ) −1 )−75 0 75Velocity (km s −1 ) −0.02−0.010.00 S i n / S c o m p − −0.0150−0.00750.00000.00750.0150 − W H α ( Å ) S i n / S c o m p − −0.0150−0.00750.00000.00750.0150 − W H α ( Å ) S i n / S c o m p − −0.0150−0.00750.00000.00750.0150 − W H α ( Å ) S i n / S c o m p − −0.0150−0.00750.00000.00750.0150 − W H α ( Å ) S i n / S c o m p − −0.0150−0.00750.00000.00750.0150 − W H α ( Å ) S i n / S c o m p − −0.0150−0.00750.00000.00750.0150 − W H α ( Å ) = η HK = Comp. spectrum −0.02−0.010.00 S i n / S c o m p − Figure 1 . H α absorption timeseries for the seven dates from Table 1. The W Hα points are shown with red circles. Observations usedas comparison spectra are shown as purple circles. All dates are shown on the same scale. The thick pink line shows the scaled andmedian corrected η HK values in order to convey how the Ca II flux changes correlate with changes in W Hα . The right column shows themaster in-transit transmission spectrum for each transit. Note that only observations showing 1 σ absorption are included in the mastertransmission spectrum. with smaller numbers of comparison spectra. The standard deviation uncertainties are ∼ √ N where N is the number of comparison spectra used. We recommend that future studies adopt the more conservative standarddeviation estimate for individual time series points. Table 2 . H α absorption and η HK values W aHα Mean in-transit − η bHK Mean comparison − η HK UT Date (10 − ˚A) (˚A) (˚A)(1) (2) (3) (4)2006 Aug 21 9.11 ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± a Uncertainties are calculated by propagating the flux uncertainty for each spectrumthrough Equation 2. b Uncertainties are the standard deviation of the points included in the mean.
In order to test for correlations between the stellar activity level and W Hα , we define the following measure of theaverage between the Ca II H and K core fluxes: η HK = 12 (cid:34)(cid:88) (cid:0) F HN − (cid:1) ∆ λ H + (cid:88) (cid:0) F KN − (cid:1) ∆ λ K (cid:35) (3)where F HN is the core flux of the Ca II H line normalized to the 0.1 ˚A wide regions centered at ± λ H is the dispersion near Ca II H and the flux integration is done between λ H ± . II K. We note that we do not perform the residual Ca II H and K core analysisfrom Cauley et al. (2016) due to the low S/N of the HARPS spectrum. The η HK measures essentially the same thingbut information about the shape of the residual line profile is lost. In-transit and comparison spectra η HK values aregiven in Table 2. The mean Ca II H and K profiles for the comparison observations are shown in Figure 2. The medianmaximum normalized flux is marked with a red dashed line.Figure 1 shows W Hα (red circles) as a function of the time from mid-transit for each of the transits listed in Table 1.All rows are on the same vertical and horizontal scale. The thick pink lines show a scaled value of η HK that hasbeen shifted by the median η HK value of the comparison spectra (purple circles). The vertical green lines show thetransit contact points and the horizontal blue line marks W Hα = 0 .
0. Consistent transits in H α are detected for theHARPS data on 2006 September 8 and 2007 July 20. Transits showing transient absorption signatures are detectedfor 2006 August 21, 2007 August 29, and 2013 June 3. The shape of the W Hα HARPS timeseries measurements arevery similar to those presented by Barnes et al. (2016).A few things are immediately evident from Figure 1. First, the H α signal is highly variable, showing large deviationsfrom epoch to epoch in both the strength of the absorption and in the duration of the transit. Second, the datafrom 2006 August 21, 2007 August 29, and 2013 June 3 show abrupt in-transit changes in W Hα . This contrasts withthe other four dates which show fairly uniform in-transit absorption that is typical of, for example, a non-varyingextended atmosphere. Lastly, 2006 August 21, 2013 July 4, and 2015 August 4 show evidence of absorption outsideof the optical transit times: 2006 August 21 shows post-transit absorption, 2013 July 4 shows pre-transit absorption,and 2015 August 4 shows both pre- and post-transit absorption. The pre-transit signals from 2013 July 4 and 2015August 4 have been explored in Cauley et al. (2015, 2016). The abrupt in-transit changes could be due to transits ofactive regions on the stellar surface or varying levels of stellar activity. Active region transits will be investigated inSection 6 and correlations with the Ca II H and K lines will be discussed in the next subsection. −75 0 750123 −75 0 750123
Ca II H −75 0 75−75 0 75 −75 0 75−75 0 75 −75 0 75−75 0 75 −75 0 75−75 0 75 −75 0 75−75 0 75 −75 0 75−75 0 75 N o r m a li z ed i n t en s i t y Velocity (km s −1 ) Figure 2 . The normalized mean Ca II H and K profiles for the comparison spectra from each date. The dashed charcoal line shows thenormalization level and the dashed red line shows the median maximum flux level for all dates. Note the strong excess core emission for2015 August 4.
Stellar activity versus planetary absorption
As discussed in Cauley et al. (2015, 2016), changes in the stellar activity from one observation to another can mimicabsorption by circumplanetary material. We note that this is separate from the in-transit contrast effect, which isdiscussed in Section 6. One method of distinguishing absorption from stellar variability is to compare the W Hα valueswith simultaneous measurements of an independent measure of the stellar activity level. Changes in W Hα can be moreconfidently attributed to absorption if there is no correlation with the independent stellar activity measure.Figure 3 shows the values of W Hα versus η HK for each epoch. In-transit observations are shown in dark red whileout-of-transit observations are shown in dark blue. We calculate Spearman’s ρ rank correlation coefficients for eachdate. The ρ S value and corresponding two-sided significance p are shown in green in the upper-right of each panel.Only the dates of 2013 June 3, 2013 July 4, and 2015 Aug 4 show significant ( p (cid:46) .
05) correlations between W Hα and η HK . The 2015 Aug 4 correlation is largely driven by the distinct in- and out-of-transit groupings, i.e., there isno correlation within the in-transit points or within the out-of-transit points. The clumping of the points from 2013June 3 likely prevents the correlation from being stronger but η HK clearly changes at the same time as W Hα near t − t mid = 0 minutes. We discuss the relationship between W Hα and η HK for the uniform W Hα transits in the nextsection.For the non-uniform transits of 2006 August 21, 2007 August 29, and 2013 June 3, the interpretation of the η HK and W Hα relationship is uncertain. The small number of comparison points and low S/N of the 2007 August 29 datamake this transit especially difficult to interpret. The 2006 August 21 and 2013 June 3 transits both feature abruptchanges in W Hα at mid-transit, producing strong absorption in the case of 2006 August 21 and filling in the absorptionin the case of 2013 June 3. The clear change in η HK at a similar time in the 2013 June 3 transit suggests that thismay be entirely attributable to changes in the stellar activity level. This is not the case for 2006 August 21.It seems unlikely that such abrupt changes can be due to physical variations in the planetary atmosphere. However,transiting gas (e.g., previously evaporated material) not associated with the atmosphere could cause such changes. Inthis case, the atmosphere would show no absorption and the external feature (e.g., a condensation or accretion stream;Lai et al. 2010; Lanza 2014) would be orbiting ahead of the planet in the case of 2013 June 3 and behind the planet forthe 2006 August 21 data. The feature would have to be ∼ R p ahead of or behind HD 189733 b for the transit to endor begin halfway through the planet’s transit. Although it is unclear as to the exact nature of these short-term W Hα variations, which are present in all of the transits, we showed in Cauley et al. (2017) that changes of the magnitudesseen, for example, in the 2013 June 3 and 2006 August 21 data, are rare when the planet is out-of-transit, suggesting −0.02 −0.01 0.00 0.010.00−W H α (Å)0.510.570.62 − η H K ρ S =0.49p=1.7e−4 − η H K ρ S =0.57p=6.5e−4 − η H K ρ S =0.50p=0.05 − η H K ρ S =−0.009p=0.97 − η H K ρ S =0.18p=0.45 − η H K ρ S =0.30p=0.22 − η H K ρ S =0.16p=0.23In−transitOut−of−transit Figure 3 . Correlation plots between W Hα and η HK . In-transit points are shown in dark red and out-of-transit points are shown in darkblue. The median in-transit W Hα value is marked with a vertical magenta dashed line. Correlations significant at the 99% level are presentfor 2007 July 20 and 2013 July 4. that they occur preferentially when the planet is in- or near-transit. We find it more likely that these changes aredue variable absorption in the circumplanetary environment since stochastic changes in the stellar activity level, as aresult of star-planet interactions, should be observable at orbital phases not associated with the transit.4.2. Ca II as an absorber Although Ca II emission is a widely examined indicator of chromospheric activity, it is possible that Ca II atoms inthe extended atmosphere of HD 189733 b might have enough opacity to absorb stellar photons (Turner et al. 2016).Lanza (2014) posited that Ca II absorption in stellar prominences fed by planetary mass loss could be responsible forthe correlation between planetary surface gravity and log R (cid:48) HK found by Hartman (2010) (see also Fossati et al. 2015).Indeed, detections of multiple transitions of neutral calcium have been claimed in the atmosphere of HD 209458 b(Astudillo-Defru & Rojo 2013). If there is significant Ca II absorption by the planet, the measured in-transit core fluxis no longer directly tracing the short-term stellar activity level, especially when comparing in-transit observations toout-of-transit observations. This is hinted at by the fact that 5 of 7 transits have lower mean in-transit η HK valuescompared to the comparison spectrum η HK value (see Table 2). We note, however, that the probability of measuring adecrease in η HK in at least 5 of the 7 transits is ∼ II measurements showa preference for absorption. Increasing the number of transits at Ca II would help clarify if the observed statistics arerepresentative of a real effect.Planetary Ca II absorption may explain the data for the dates showing lower in-transit − η HK values and fairlysmooth W Hα transits. For the 2006 September 8, 2007 July 20, 2013 July 4, and 2015 August 4 transits the mean η HK value for the comparison spectra is more negative (meaning more core flux) than the mean in-transit η HK value.We find significant correlations between W Hα and η HK for 2013 July 4 and 2015 Aug 4. The 2013 July 4 correlation isdriven by a correlation within the out-of-transit points ( ρ S = 0 . p = 0 . W Hα and η HK from one exposure to another. This tentatively suggeststhat the lower in-transit η HK differences are not due to changes in the stellar activity level.An alternative to absorption in the planetary atmosphere producing the lower in-transit η HK values is the transitingof bright active regions (Llama et al. 2015). If the planet happens to transit an especially active latitudinal bandthat is bright in Ca II compared to the rest of the star, the Ca II core emission will be preferentially blocked by theplanet’s disk, resulting in lower in-transit η HK values. Unless the active latitude is fairly uniform in brightness, thetransit tends to be choppy and uneven (Llama et al. 2015). Although the Ca II light curves in Figure 1 do show somesignificant exposure-to-exposure variations, the epochs with lower in-transit versus out-of-transit η HK values (e.g.,2007 July 20 and 2013 July 4) exhibit fairly smooth time series. Thus we tentatively favor the atmospheric absorptionscenario over active region transits, although a thorough investigation of the contrast effect at Ca II H and K wouldbe informative.If Ca II in the planetary atmosphere is indeed absorbing, it cannot be considered a reliable independent measure ofthe stellar activity level near (i.e., immediately pre- or post-transit) or during the planet’s transit. Detailed theoreticalmodeling of the Ca II population in the extended atmospheres of hot exoplanets would be useful in determining ifCa II should be specifically targeted in these systems. A possible alternative is simultaneous monitoring of a FUVactivity diagnostics (e.g., France et al. 2016) that are not expected to be present in the extended atmospheres of hotplanets. EPOCH TO EPOCH VARIATIONSThe significant exospheric changes measured in Ly α by Lecavelier des Etangs et al. (2012) and Bourrier et al. (2013)for HD 189733 b motivates us to look for changes in the H α signal as a function of stellar activity level. This hasalso been investigated previously by Barnes et al. (2016) for the HARPS transits. Christie et al. (2013) showed thatthe strength of H α planetary absorption depends strongly on the Ly α and ionizing flux from the star. In order tomatch the strength of H α absorption measured by Jensen et al. (2012), Christie et al. (2013) find that a large value ofthe ionizing flux is needed. For reference, the H α absorption measured by Jensen et al. (2012) is comparable to thatmeasured for the 2015 August 4 transit. This result suggests that dates showing larger values of out-of-transit η HK should show larger amounts of in-transit H α absorption.The strongest core emission in Figure 2 is clearly seen for 2015 August 4. Figure 4 shows the mean − η HK value fromFigure 2 plotted against the mean in-transit − W Hα from each date. The uncertainty in each value is the standard0 −0.014 −0.012 −0.010 −0.008 −0.006Mean in−transit −W H α (Å)0.20.30.40.50.6 M ean c o m pa r i s on − η H K ( Å ) −0.014 −0.012 −0.010 −0.008 −0.006Mean in−transit −W H α (Å)0.20.30.40.50.6 M ean c o m pa r i s on − η H K ( Å ) Figure 4 . Comparison between the mean in-transit values of W Hα and − η HK calculated for the mean Ca II H and K comparison spectrashown in Figure 2. Uncertainties in η HK for 2013 June 3 are smaller than the size of the plot symbol. There is no clear trend. However,the date with the highest η HK by a large margin is 2015 August 4 (purple) and this date also shows the deepest and most extended H α transit, suggesting that the extended atmosphere is strongly influenced by the stellar activity level if the H α signal arises in the planetaryatmosphere. deviation of the points included in the mean value. Only points showing 1 σ significant absorption are included in themean W Hα values.There is no clear trend between W Hα and η HK from epoch to epoch. However, 2015 August 4 stands out: it showsthe strongest in-transit H α absorption signal and the largest out-of-transit η HK value. In addition, the 2015 August 4H α transit appears significantly extended both pre- and post-transit, hinting at an atmosphere that fills a significantfraction of the planet’s Roche lobe. The low H β and H γ absorption values from Cauley et al. (2016) also supportthe presence of a low-density extended atmosphere. This suggests that there may be some threshold (e.g., − η HK (cid:38) η HK within a single night are much less than the inter-epoch variations.Thus although we have argued that Ca II may be absorbing stellar photons around the planet, we do not expect theabsorption effect to be as large as night to night variations in the stellar activity level, as seen, for example, by Boisseet al. (2009) across many weeks of H α and Ca II observations of HD 189733 and similarly at H α by Cauley et al.(2017).It is not clear if the star was in a particularly active state during the 2015 August 4 transit or an especiallyactive hemisphere of the star was visible during that night. The suggested threshold needs to be confirmed with alarger sample of transits, ideally during periods when the star is in an especially active state, that show consistentH α absorption since the interpretation of irregular transits (e.g., 2006 August 21 and 2007 August 29) is much lessstraightforward.Overall, we find no strong evidence for a relationship between the stellar activity level and the strength of thein-transit W Hα measurement, although the date with the highest η HK value also shows the strongest in-transit H α absorption. The analysis presented here should be supplemented in the future with more transit observations, prefer-ably containing as many simultaneous activity indicators as possible. In the next section, we investigate whether ornot the contrast effect for HD 189733 can explain the observed H α absorption signatures. A similar analysis will bepresented in a future paper for Ca II , Na I , and Mg I . THE CONTRAST EFFECT FOR HD 189733While much effort has been made to calculate the effects of star spots, faculae and plage , and other irregular surfacefeatures on stellar radial velocities, broadband transmission spectra, and properties derived from transit measurements We do not distinguish between faculae, which form in the photosphere, and plages, which form in the chromosphere. The spectra usedto model bright H α regions on the stellar surface are a combination of both plage and facular characteristics. α contrast effect, i.e.,how the ratio of in-transit to out-of-transit flux as a function of wavelength changes based on which portion of thestar is being occulted by the planet. This is critical to understanding how the surface of an active star such as HD189733 will affect the W Hα measurement, especially in light of the recent suggestion by Barnes et al. (2016) that theH α signal is due mostly to variations on the stellar surface and not to absorption by circumplanetary material.6.1. Model overview
Here we provide a brief overview of the steps involved in creating the model transmission spectra. The stellar andplanetary parameters used in the model are given in Table 3. The model has eleven input parameters which are listedin Table 4. Table 4 also shows the range of parameter values we explored with the contrast model but most of theexamples presented below focus on a much smaller range or even specific values. This is due to the fact that someparameters have little influence on the contrast effect.We first simulate the location of spots and faculae on the stellar disk. We assume that each spot has some concentricassociated facular region that has an area equal to 60% of the spot area. After all spot locations have been determined,additional facular regions are added to match the desired spot-to-facular ratio Q = A sp /A fac . The location of a spotis determined by randomly selecting from a normal distribution in latitude centered on the latitude input parameter.The FWHM of the distribution is kept fixed at 10 ◦ . The size of the spot is then determined by randomly selectingfrom a uniform distribution with boundaries specified by r minsp and r maxsp , the minimum and maximum allowed spotradii in units of R p . We do not allow overlap of spots. Finally, the projected boundaries, which are ellipses, of the spotand surrounding faculae are determined based on the central spot location. The additional facular regions are addedin an identical procedure except the edges of the ellipses may overlap to allow potentially more irregular patterns. Table 3 . Stellar and planetary parameters
Parameter Symbol Value a Units(1) (2) (3) (4)Stellar radius R ∗ R (cid:12) Stellar rotational velocity v sin i − Impact parameters b R ∗ Orbital period P orb v orb − Orbital semi-major axis a R p R J a With the exception of v sin i , all stellar and planetary parameterstaken from Torres et al. (2008). The v sin i value is taken fromCollier Cameron et al. (2010). Table 4 . Contrast model parameters and explored values
Parameter description Symbol Value range(1) (2) (3)Spot coverage fraction A sp Q A fil Table 4 continued on next page Table 4 (continued)
Parameter description Symbol Value range(1) (2) (3)Ratio of facular to photosphere core H α q fac σ fac − Central latitude of active region distribution θ act ◦ -45 ◦ Temperature difference between spots and photosphere ∆ T sp T fac α core contrast c fil r minsp R p Maximum spot radius r maxsp R p We also include filaments which appear in absorption against the stellar disk (e.g., Heinzel & Anzer 2006; Kuckeinet al. 2016). The filaments have a fixed width of 0.2 R p and their length is defined by the total filament coveragefraction. Filaments are constructed by choosing a random starting point and direction and then letting the filamentgrow in that direction with a narrow cone defining the directions each new piece of the filament is allowed to movetowards. The filament is allowed to grow until the desired coverage fraction is achieved to within 0.1%.To simulate the transmission spectrum, we first calculate the out-of-transit spectrum by summing the spectra fromindividual grid points across the entire stellar surface. The grid resolution is 0.02 R p which results in a 647 × α transmission spectra. In addition, the exact line shape is less important than themagnitude of the effect, which will not change significantly if a Lorentzian or Voight profile is assumed. The spectrumat each point on the stellar disk is shifted by the appropriate stellar rotational velocity, for which we assume rigidrotation. We also compute the radial velocity of the star relative to the observer (see Eq. 11 of Lovis & Fischer 2010)and apply this velocity shift to each of the stellar spectra. We then calculate the stellar spectrum blocked by theplanet at 5 minute intervals across the transit. The out-of-transit spectrum is subtracted from the blocked spectrumand then the result is divided by the out-of-transit spectrum to generate synthetic transmission spectra according toEquation 1. An example of the stellar surface for the parameters A sp = 0 . Q =0.7, and A fil =0.005 is shown inFigure 5. 6.2. Surface feature spectra
Each distinct surface feature contributes a different H α spectrum. We include four separate surface components: 1.the naked photosphere; 2. star spots; 3. faculae; and 4. filaments. Below we discuss each spectrum in detail, as well asthe other parameters associated with that specific surface feature. Example H α spectra of the four surface componentsat µ = cos θ = 1 .
0, where θ is the angle between the normal vector to the stellar surface and the line-of-sight, are shownin Figure 6 for a model with T sp =4300 K, q fac =1.5, T fac =5040 K, and c fil =0.6. The faculae are approximately thesame brightness as the photosphere at µ =1.0 (see Equation 4). The normalized continuum flux for the spot spectrumis reduced by a factor of ( T sp /T eff ) . We ignore all small-scale velocity effects in the spectra, such as photosphericconvective blue shifts, since we are only concerned with the overall contrast between the different regions. The contrastis weakly affected by these velocity shifts, which are of the order ∼ − (e.g., Meunier et al. 2010; Lanza etal. 2011). 6.2.1. Photospheric spectra
For the photosphere we use a PHOENIX model spectrum (Husser et al. 2013) with T eff =5000 K, log g =4.5, and[Fe/H]=0.0. It is important to emphasize that the choice of photospheric spectrum is not critical to exploring thecontrast effect since most of the stellar surface is dominated by the photosphere. Furthermore, the spotted, filament,and facular spectra are scaled by or built from the photospheric spectrum in some manner so the entire surface isconstructed relative to the photosphere. Thus a choice of the model T eff of 4500 K or 5500 K does not significantlyaffect our conclusions. We note that Boyajian et al. (2015) recently measured T eff for HD 189733 to be 125 K lower(4875 ±
43 K) than the value used here.3 −6 −4 −2 0 2 4 6x (R p )−6−4−20246 z ( R p ) Figure 5 . Example stellar disk for a contrast model with A sp = 0 . Q =0.7, and A fil =0.005. The faculae are shown as white annulisurrounding the dark brown spots and randomly placed circles with the same latitudinal distribution as the spots. A dark filament canbe seen across disk center. The transit path of HD 189733 b is shown by the dashed gray lines. Note that the limb-darkening, and limb-brightening in the case of the faculae, is meant to be representative of how the features contribute near the H α line core (i.e., absorptionor emission) and not to show the precise intensity ratios calculated as a function of wavelength in the models. −50 0 50Velocity (km s −1 )0.00.20.40.60.81.01.2 N o r m a li z ed f l u x PhotosphereSpotFaculaeFilament Figure 6 . Model surface feature H α spectra at µ = 1 for T sp =4300 K, q fac =1.5, T fac =5040 K, and c fil =0.6. Each spectrum is normalizedby fitting a line to the flux at −
200 and +200 km s − . Note that these are spectra of individual surface elements and do not represent theabsolute fluxes contributing to the explored contrast examples. The core emission in the facular spectrum (purple line) completely fills inthe photospheric absorption in this case. We note that this is more typical of a flare rather than a quiescent facular region and is shownhere to illustrate the differences between the spectra. The temperature difference between the spot and the photosphere results in ∼ α core absorption superimposed on the photosphericspectrum. The narrow feature near 30 km s − is a Cr I line. We ignore the absorption contribution from this line when calculating W Hα . Spot spectra
For the star spot spectra we use PHOENIX model spectra with the same log g and [Fe/H] values but with varyingtemperatures in step sizes of 100 K. Spots are cooler than the surrounding photosphere with differences in temperaturethat depend on spectral type, although this dependence may break down at T eff > T (cid:12) (Eker et al. 2003; Berdyugina2005). We note that we do not distinguish between umbral and penumbral temperatures; they are averaged into asingle spot temperature. Herrero et al. (2016) demonstrate (their Figure 2) the good agreement between observedsolar spot spectra and model PHOENIX spectra at a cooler T eff .Pont et al. (2013) provided a detailed analysis of spot crossing events during HD 189733 b transits. Their mainconclusions were: 1. there is no evidence for the transiting of concentrated bright faculae, suggesting that facularregions are spread out across the stellar surface; 2. the spot filling factor is ∼ T = 750 K. Photometric modeling performed by Herrero et al. (2016) suggeststhat spots dominate changes in the stellar brightness across stellar rotation periods. The observations presented inPont et al. (2013) span ∼ Filament spectra
Filament spectra are identical to photospheric spectra but with the H α line core subject to further absorption.Kuckein et al. (2016) measure the contrast across H α from ∼−
80 to +80 km s − for filaments on the solar disk. Theyfind ∼
60% less flux in the line core compared with the bare photosphere, although the values vary depending on thewhere the measurement is made on the filament. We adopt a single contrast value of 0.6 (i.e., 60% less flux) at the linecore for the filament spectra. The absorption is modeled as a Gaussian with a full width at half maximum (FWHM)of 40 km s − . We do not consider prominences, which are filaments projected beyond the edge of the stellar disk,although large prominences may produce pre- or post-transit contrast signals.6.2.4. Facular spectra
Facular spectra consist of the intensity-weighted sum of the underlying photosphere and overlying emission region.During small solar flares, the ratio of H α line core emission to the underlying photospheric spectrum can reach ∼ ∼ ∼ q fac = 0 . − . q fac ∼ . − .
0. Larger ratios than this likely do not apply to non-flaring regions. However,we explore larger values of q fac since larger values are needed to reproduce the observed W Hα values. Whether or notthese large values for the H α core emission are physical will be discussed below.Facular and plage regions are limb-brightened relative to the underlying photosphere. We adopt the limb-brighteninglaw of Meunier et al. (2010) and Herrero et al. (2016): c fac ( µ ) = (cid:18) T eff + ∆ T ( µ ) T eff + ∆ T fac (cid:19) (4)where ∆ T fac is the temperature difference between the photosphere and faculae and ∆ T ( µ ) = 250 . − . µ +190 . µ .6.3. Contrast and limb-darkening/brightening
Czesla et al. (2015) calculated differential center-to-limb variations (CLV), or differences in the limb-darkening orbrightening as a function of wavelength across a specific spectral line, for the Ca II H and K and Na I D lines forHD 189733. They demonstrated that the Na I lines show limb- brightening in the wings of the line compared with theline core or neighboring continuum. This is important since a transiting planet with no atmosphere can produce atransmission signal, even if there are no active regions present on the stellar surface, due to the fact that the ratio ofthe line core to the continuum changes as a function of the transit (see Figures 1-5 of Czesla et al. 2015). This effectwas recently included in Na I transit modeling by Khalafinejad et al. (2016).Such CLVs will also affect the in-transit H α measurements. In order to account for this effect, we have calculated high-resolution H α spectra using the program SPECTRUM by Gray & Corbally (1994) and an ATLAS9 model atmosphere T eff =5000 K, log g =4.5, and [Fe/H]=0.0. The spectra were computed at fourteen values of µ = cos ( θ ) between0.01 and 1.0. To avoid interpolation during the contrast model calculations, we fit the following limb-darkening lawfrom Hestroffer & Magnan (1998) to each wavelength across the spectrum: I ( µ ) = 1 − u (1 − µ α ) (5)The parameters from Equation 5 are then used to compute I ( µ ) as a function of wavelength across the line for adensely sampled grid of µ across the stellar disk. Examples of the I ( µ )/ I ( µ = 1) vs. µ curves and the correspondingEquation 5 fits are shown in Figure 7. There is a significant difference between the limb-darkening in the core of theline versus the line wing which can impact the measured W Hα values. µ I ( µ ) /I ( µ = ) Line centerLine wing (−2.2 Å)Eq. 5 fits µ I ( µ ) /I ( µ = ) Figure 7 . Examples of wavelength-dependent limb-darkening curves from the synthetic spectra (solid lines and symbols) and the fits usingEquation 5 (red dashed lines). Note the steeper limb-darkening in the line core compared with the line wing.
Figure 8 shows a transit example for a pure photosphere, i.e., no spots, faculae, or filaments. The only mechanismaffecting the transmission spectrum, or, in the absence of a planetary atmosphere, the contrast spectrum, is the CLVsdescribed above. Five representative in-transit times are shown in the left panel of Figure 8 and their corresponding S T profiles are shown in the middle panel. The deepest contrast profiles are seen when the planet is near the stellarlimb, but still almost completely covers the disk, since this is where the limb-darkening curves in the line wing andline core differ the most. The W Hα values for the entire in-transit calculation are shown in the right panel of Figure 8which explicitly shows the “absorption” effect near the stellar limb.An important takeaway from the example shown in Figure 8 is that the signal induced by CLVs at H α is well belowthe measured W Hα values for most of the transit presented here. This is especially true for values of W Hα near mid-transit where the CLV effect is weakest. Although HD 189733 is an active star and thus it is unlikely that the visiblehemisphere is ever pure photosphere, this baseline demonstrates that the observed H α transit signals cannot be causedby only CLVs in this specific case. However, the magnitude of the CLV effect is significant upon ingress and egressand must be included in any model of the in-transit absorption. It is also important to highlight the velocities of the S T profiles in the middle panel of Figure 8: upon ingress, S T shows red-shifted velocities near maximum absorptionwhile upon egress it shows blue-shifted velocities. This contrast signal will have an important effect on the v Hα modelspresented in Section 7. 6.4. A note on parameter importance
Specifying unique spectra and a unique spatial configuration for active regions on the stellar disk requires all ofthe parameters in Table 4. However, some of these parameters are much more important than others in determiningchanges in the contrast spectra. In addition, we can look to the observed H α transmission spectra and previous HD189733 studies for guidance on other parameter values. For example, the spot radii actually govern the distributionof spots since larger spot radii means fewer spots for a constant A sp . But A sp is more important for determining6 −6 −4 −2 0 2 4 6x (R p )−6−4−20246 z ( R p ) −54 −36 −18 0 18 36 54t−t mid (minutes) −100 −50 0 50 100Velocity (km s −1 )−0.004−0.0020.0000.0020.004 S T −60 −40 −20 0 20 40 60t−t mid (minutes)−0.003−0.002−0.0010.0000.001 − W H α ( Å ) Figure 8 . Example of the contrast effect for a star with no spots, faculae, or filaments. The planet’s position at five in-transit times isshown with the solid circles (left panel). The top x-axis gives the time from mid-transit corresponding to the distances on the bottomx-axis. The transmission spectra S T for each in-transit time are shown in the middle panel. The narrow spikes near 27 km s − are theresult of a Cr I line at 6563.40 ˚A. Note that this line is ignored in the calculation of W Hα . S T shows a shift from stronger to weakerabsorption as the planet approaches mid-transit. This is the result of CLVs where the difference between the limb-darkening in the wingversus the line core is smallest near the middle of the stellar disk. Note the difference in scale between the right panel and the transmissionspectra shown in Figure 1. The CLV effects are an order of magnitude lower than the measured signals. The right panel shows the W Hα values calculated for the full set of in-transit times. The vertical dashed lines mark the transit contact points and the vertical solid linemarks mid-transit. The strongest absorption signal, or in this case contrast signal, is when the planet is near the edge of the stellar diskbut still almost completely covers the limb. This is very similar behavior to the Na I D calculations presented by Czesla et al. (2015). the relative weights between spotted spectra and the photosphere, which in turn is more important for the contrastspectrum. The distribution of active regions is a strong determinant of the time series but does not strongly affect thecontrast spectrum if the planet is not occulting a large fraction of the active regions. Another example is σ fac of theactive region emission. With the assumption of the spectra being produced by the contrast effect, we can infer that σ fac ≈
40 km s − . This suggests that we do not need to explore a large range of values for σ fac since most valueswill not be able to reproduce the line profile shape. For these reasons, although we have investigated the full range ofparameters given in Table 4, we do not present details of the full extent these efforts and instead focus on illuminatingcases that are most relevant to the measurements.6.5. Active stellar surface transits
In this section we present various illustrative examples of the contrast effect for an active stellar surface. Theseexamples are not meant to be exhaustive but rather representative of how general configurations affect the contrastspectrum and which parameters are most important in producing a significant contrast spectrum. We do not explorescenarios where the active regions are isolated to a small portion of the stellar disk since transits of these isolatedregions have short durations. Instead, we explore scenarios that might produce absorption across the entire transit.6.5.1.
Transiting a photospheric chord
The planet can either transit active regions or the photosphere. Either way, the in-transit spectrum will be alteredrelative to the out-of-transit spectrum. Figure 9 shows an example of the planet transiting a chord with no activeregions for the parameter values A sp = 0 . Q = 0 . A fil = 0 . θ act = 10 ◦ , and q fac = 1 .
5. The CLVs can still beseen in the contrast profiles (middle panel) but since the in-transit spectra are weighted towards active regions, S T shows emission in the line core at most t − t mid rather than absorption. This also shifts the − W Hα curve up towardsless absorption.One important thing to reiterate in this example is that strong absorption lines are not produced in the contrastspectra. This occurs for two reasons: 1. the facular core emission dominates the contrast spectrum; 2. the spotspectrum actually has a smaller core-to-continuum contrast compared with the photospheric spectrum . This reflects thefact that the H α line strength decreases for cooler stellar spectra. In this example, ∆ T sp = 700 K so the spot spectrumis identical to the spot spectrum in Figure 6. This has important consequences for the pure-spot scenario, which wedo not show: even without the core emission from the facular/plage regions, the contrast spectrum would still show7 −6 −4 −2 0 2 4 6x (R p )−6−4−20246 z ( R p ) −54 −36 −18 0 18 36 54t−t mid (minutes) −100 −50 0 50 100Velocity (km s −1 )−0.020−0.015−0.010−0.0050.0000.005 S T −60 −40 −20 0 20 40 60t−t mid (minutes)−0.020−0.015−0.010−0.0050.0000.005 − W H α ( Å ) Figure 9 . Transit of a photospheric chord with A sp = 0 . Q = 0 . A fil = 0 . θ act = ± ◦ , and q fac = 1 .
5. The 2013 Jul 4 and 2015Aug 4 Keck data are shown in green in both the middle and right panels. Note the different scale compared with Figure 8. Since thein-transit spectrum is weighted towards the active regions, the W Hα curve (right panel) is shifted up due to the line core being filled in byfacular emission. This can also be seen in S T (middle panel) where the spectra show stronger emission features. −6 −4 −2 0 2 4 6x (R p )−6−4−20246 z ( R p ) −54 −36 −18 0 18 36 54t−t mid (minutes) −100 −50 0 50 100Velocity (km s −1 )−0.020−0.015−0.010−0.0050.0000.005 S T −60 −40 −20 0 20 40 60t−t mid (minutes)−0.020−0.015−0.010−0.0050.0000.005 − W H α ( Å ) Figure 10 . Same as Figure 9 but for the pure filament case with A fil = 0 .
01. The filaments produce a very weak contrast effect. core emission instead of absorption since the spot spectrum shows less core absorption than the photosphere. As aresult, spots have little effect on the H α transmission spectrum.We also show the filament-only case in Figure 10 for A fil = 0 .
01. Filaments do not produce strong contrast spectrafor reasonable values of A fil , although we note that the average transmission spectrum (not shown) for an entiretransit is seen in weak absorption for the case of the photospheric chord transit. Individual transmission spectra areseen in emission for transits of filaments. Since their contribution to the contrast effect is minor, even when the planettransits filaments, we do not focus on them further.6.5.2. Transiting an active latitude
The photospheric transit examples demonstrate that W Hα cannot be measured in absorption if the planet does notconsistently occult facular regions. Thus in order to produce the strongest W Hα curves shown in Figure 1 HD 189733b must be transiting a chord that is densely covered with faculae and plage regions. Furthermore, as we demonstratebelow, the emission strength in these facular regions must be similar to what is seen in flaring regions on the Sun.Here we present examples of the planet transiting chords that contain active regions. The first scenario is shown in8Figure 11 for uniformly distributed active regions with A sp = 0 . Q = 0 . A fil = 0 .
0, and q fac = 4 .
0. The W Hα timeseries in the right panel is erratic; there is no trend. We have chosen the large value of q fac = 4 . q fac : in Figure 12 q fac = 0 .
5, while in Figure 13 q fac = 1 . q fac the absorption line is ∼ × stronger. Furthermore, the W Hα values consistently show absorption across the entire transit whereas in the q fac = 0 . W Hα approacheszero near mid-transit. These examples illustrate the important point that the strength of q fac is the dominant factorin determining the magnitude of the contrast effect for any given transit snapshot. For the W Hα timeseries, Q , andtherefore the facular coverage fraction, is also important. But q fac is the main determinant of the depth of the contrastspectrum. In neither case does the contrast effect produce absorption similar to what is observed. −6 −4 −2 0 2 4 6x (R p )−6−4−20246 z ( R p ) −54 −36 −18 0 18 36 54t−t mid (minutes) −100 −50 0 50 100Velocity (km s −1 )−0.020−0.015−0.010−0.0050.0000.005 S T −60 −40 −20 0 20 40 60t−t mid (minutes)−0.020−0.015−0.010−0.0050.0000.005 − W H α ( Å ) Figure 11 . Same as Figure 9 but now with A sp = 0 . Q = 0 . A fil = 0 .
0, and q fac = 4 . W Hα (right panel) is erratic since the planet never occults a large area of active regions. Note that q fac is very large inthis example but very similar behavior is seen for smaller values, although the variations in W Hα are smaller. The previous two examples demonstrate the need for greater contrast at H α in order to reproduce the observations.Figure 14 and Figure 15 show scenarios with very high surface coverage fractions ( A sp = 0 .
03 and Q = 0 .
15 in bothcases) and very strong facular/plage emission ( q fac = 4 . θ act = ± ◦ while in Figure 15 it is centered at θ act = ± ◦ . In Figure 14 the planettransits an almost constant area of active regions and the strength of the facular emission produces strong contrastabsorption lines (center panel) and a relatively uniform W Hα timeseries (right panel). Although q fac is the same inFigure 15 the contrast profiles are weaker and W Hα is ∼ × shallower. This is the result of the planet transiting theedge of the active region distribution where, at any in-transit snapshot, the planet occults a relatively smaller area offacular/plage regions compared to the distribution in Figure 14. This suggests that the active regions on HD 189733’ssurface must be centered very near the planet’s transit chord in order to produce a relatively uniform W Hα timeseries. The value of Q in the above examples, which equates to a 20% surface coverage of facular/plage regions, does notnecessarily need to be so low, i.e., the active region surface coverage does not need to be so large. Figure 16 shows thecase for A sp = 0 . Q = 0 . A fil = 0 .
0, and q fac = 4 .
0. While W Hα is not as uniform, the values during the firsthalf of the transit approach the largest observed values from Figure 1. We note that moving the center of the activeregion distribution to ± ◦ latitude results in essentially no contrast for the entire transit. This again illustrates theneed for the active regions to be concentrated near the transit chord and to be uniformly distributed in longitude. Italso shows that if Q increases, q fac must remain high or even increase in order to get close to the observed values;decreasing q fac from 4.0 to ∼ −6 −4 −2 0 2 4 6x (R p )−6−4−20246 z ( R p ) −54 −36 −18 0 18 36 54t−t mid (minutes) −100 −50 0 50 100Velocity (km s −1 )−0.020−0.015−0.010−0.0050.0000.005 S T −60 −40 −20 0 20 40 60t−t mid (minutes)−0.020−0.015−0.010−0.0050.0000.005 − W H α ( Å ) Figure 12 . Same as Figure 9 but now with A sp = 0 . Q = 0 . A fil = 0 . θ act = ± ◦ , and q fac = 0 .
5. Now the in-transit spectrumis weighted towards the photosphere since the planet occults active regions almost continuously throughout the transit. The contrast isstrongest when the highest active region area is occulted, e.g., during the purple and red planet positions in the first panel. The W Hα values show absorption, although there is an abrupt shift near mid-transit when the planet begins to occult more of the facular regions. −6 −4 −2 0 2 4 6x (R p )−6−4−20246 z ( R p ) −54 −36 −18 0 18 36 54t−t mid (minutes) −100 −50 0 50 100Velocity (km s −1 )−0.020−0.015−0.010−0.0050.0000.005 S T −60 −40 −20 0 20 40 60t−t mid (minutes)−0.020−0.015−0.010−0.0050.0000.005 − W H α ( Å ) Figure 13 . Same as Figure 12 but now with q fac = 1 .
5. The stellar surface in the first panel is identical to that in Figure 12. The W Hα values show continuous absorption, although there is an abrupt shift near mid-transit when the planet begins to occult more of the facularregions. Although the contrast spectrum shows absorption throughout the transit, W Hα is still 2-3 × smaller than is measured in most ofthe full transits from Figure 1. Comparisons with low-activity templates
We demonstrated above that it is possible under some conditions to reproduce even the strongest observed W Hα in-transit signals using the contrast effect. However, the only parameter configurations that are able to produce these W Hα values involve modest to large facular coverage fractions and q fac ∼ −
5. Although we do not know exactlyhow the active regions are distributed on HD 189733 or the strength of the facular/plage H α core emission, we canattempt to roughly constrain the combination of these parameters by comparing HD 189733 with less active mainsequence templates of similar T eff .We have downloaded archived Keck HIRES data for the three template stars listed in Table 5. HD 189733 is alsolisted for comparison. We use these templates to find parameter combinations of q fac and A fac , where A fac is thefractional surface area covered by facular regions, that can fill in the H α absorption of the less-active template andmatch the line profile of HD 189733. Note that A fac is just another way of specifying Q in the contrast models, where Q = A sp /A fac . The HD 189733 spectrum is an average of the comparison spectra used in all of the Keck transits0 −6 −4 −2 0 2 4 6x (R p )−6−4−20246 z ( R p ) −54 −36 −18 0 18 36 54t−t mid (minutes) −100 −50 0 50 100Velocity (km s −1 )−0.020−0.015−0.010−0.0050.0000.005 S T −60 −40 −20 0 20 40 60t−t mid (minutes)−0.020−0.015−0.010−0.0050.0000.005 − W H α ( Å ) Figure 14 . Same as Figure 8 but now with A sp = 0 . Q = 0 . A fil = 0 . θ act = 40 ◦ , and q fac = 4 .
0. In this case facular/plageregions cover 20% of the stellar disk. The W Hα values show continuous absorption that is of comparable strength to the full W Hα transitsseen in Figure 1. −6 −4 −2 0 2 4 6x (R p )−6−4−20246 z ( R p ) −54 −36 −18 0 18 36 54t−t mid (minutes) −100 −50 0 50 100Velocity (km s −1 )−0.020−0.015−0.010−0.0050.0000.005 S T −60 −40 −20 0 20 40 60t−t mid (minutes)−0.020−0.015−0.010−0.0050.0000.005 − W H α ( Å ) Figure 15 . Same as Figure 14 but with θ act = 30 ◦ , and q fac = 4 .
0. The W Hα values are much weaker since the planet now transits theedge of the active region latitudinal belt. The W Hα values in the right panel are shown on the same scale as Figure 14 to emphasize themuch weaker contrast effect. −6 −4 −2 0 2 4 6x (R p )−6−4−20246 z ( R p ) −54 −36 −18 0 18 36 54t−t mid (minutes) −100 −50 0 50 100Velocity (km s −1 )−0.020−0.015−0.010−0.0050.0000.005 S T −60 −40 −20 0 20 40 60t−t mid (minutes)−0.020−0.015−0.010−0.0050.0000.005 − W H α ( Å ) Figure 16 . The less active case of A sp = 0 . Q = 0 . A fil = 0 .
0, and q fac = 4 .
0. Although the facular/plage coverage is much lowerthan in Figure 14 W Hα is consistently observed in absorption across the whole transit. S fac is constructed in the same manner as the contrast modelfacular spectrum except now q fac refers to the core flux of the template spectrum. The final spectrum is the weightedaverage of the photospheric spectrum S phot , which is the observed template spectrum, and S fac : S tot = (1 − A fac ) S phot + A fac S fac (6) Table 5 . Less-active comparison stars T eff M ∗ R ∗ ID (K) ( M (cid:12) ) ( R (cid:12) ) S HK (1) (2) (3) (4) (5)HD 189733 5040 0.81 0.76 0.51HD 192263 4975 0.83 0.75 0.47HD 104067 4956 0.91 0.75 0.34HD 87883 4958 0.78 0.77 0.28 Note —HD189733 parameters taken from Torres etal. (2008). All template stellar parameters takenfrom Valenti & Fischer (2005). Values of S HK aretaken from Isaacson & Fischer (2010). While all of the observed template spectra are rotationally broadened according to the v sin i value of HD 189733,we do not account for the CLVs discussed previously. Thus we are only exploring first-order approximations for which q fac and A fac values are needed to reproduce the HD 189733 H α core. Since we do not have spatially resolved spectraof these stars we cannot build up the spectrum across the stellar disk as is done in the contrast model case.Three different parameter combinations are shown for each template in Figure 17. A χ minimization routine is usedto produce the fits for various initial parameter combinations that correspond roughly to parameter space exploredin each column. The first column shows a direct comparison between HD 189733 (dark gray line) and the templates(orange lines). Column 2 shows the case of high A fac and low q fac , i.e., weak facular emission but distributed acrossmuch of the stellar disk. Column 3 shows slightly stronger q fac but reduced A fac . Finally, column 4 shows large q fac and small A fac . We note that the HD 189733 spectrum is also plotted in columns 2-4 but is obscured by the modelfits.For all three templates, a crucial trend is seen: as q fac increases, A fac must decrease for a given photosphericspectrum. This is not surprising but has important consequences for which parameter combinations in the contrastmodel are likely to be representative of HD 189733’s active surface. For example, the very low Q and very high q fac case shown in Figure 14 is unlikely to accurately represent the stellar surface since the same parameter combinationwould drastically overestimate the line core emission in Figure 17. However, the less active case in Figure 16 seemsplausible: similar values of q fac and Q , or a facular coverage fraction of ∼ A fac , high q fac casefor HD 87883 in Figure 17.One important caveat to the template comparison is that the template spectra themselves contain some contributionfrom active regions. Thus we are not comparing pure photospheres to HD 189733’s active surface. However, this doesnot affect the relationship between q fac and A fac ; one must decrease as the other increases for a specific templatespectrum, pure photosphere or not. On the other hand, the value of the parameters in Figure 17 may be affected. Forexample, instead of A fac = 0 .
08 and q fac = 4 .
06 in the lower right panel of Figure 17, one or both values would needto be larger if HD 87883’s spectrum contained no contributions from active regions. Thus the parameter combinationsshown in Figure 17 are likely lower limits to the true values.6.7.
Discussion and summary of contrast results
We have presented transit models for a planet with no atmosphere in order to explore the contrast effect in H α thatis produced when in-transit spectra, for which a portion of the stellar disk is occulted, are compared with out-of-transitspectra. We find the following:2 N o r m a li z ed f l u x HD 192263 q fac =0.16 A fac =0.67 High A fac , low q fac q fac =0.33 A fac =0.34 Similar A fac , q fac q fac =0.77 A fac =0.14 Low A fac , high q fac N o r m a li z ed f l u x HD 104067 q fac =0.35 A fac =0.70 q fac =0.54 A fac =0.46 q fac =4.31 A fac =0.05 N o r m a li z ed f l u x HD 87883 6560 6565Wavelength (Å) q fac =0.40 A fac =0.81 q fac =0.63 A fac =0.52 q fac =4.06 A fac =0.08 Figure 17 . Comparison of HD 189733 H α profile with less active main-sequence templates. Column 1 shows the observed templateover-plotted in orange on top of the HD 189733 spectrum (dark gray line). Columns 2-4 show various combinations of A fac and q fac from Equation 6 that are needed to reproduce the HD 189733 H α core depth. The photospheric spectrum is shown in purple and thefacular/plage spectrum is shown in green. A clear relationship is seen for A fac and q fac : as q fac increases, A fac must decrease to producethe same disk-integrated spectrum. This implies that we cannot use both high A fac and high q fac to describe the observed H α transitssince it would violate the observed disk-integrated core strength. • Spots and filaments, for the physically reasonable surface coverage fractions and spot/filament parameters ex-plored here, are unimportant in the H α contrast spectrum. The main contribution to the contrast effect comesfrom strong facular or plage emission. • Transits of the photosphere do not produce H α contrast in absorption; active regions that include strong facu-lae/plage emission must be transited to produce S T in absorption. • The facular coverage fraction must be (cid:38) −
10% and these facular regions must be concentrated around thetransit chord in order for the strongest observed H α transits to be reproduced. Large coverage fractions, nomatter the value of q fac , cannot reproduce the observed H α line profiles if the distribution is uniform across thestellar disk. This holds true even for very large coverage fractions of >
50% since the contrast spectrum thenbegins to be weighted back towards the active region spectra, producing emission instead of absorption. • Certain configurations of facular regions combined with values of q fac ∼ . W Hα ≈ . − . • The comparison of less active template stars to HD 189733 suggests that for q fac ∼ (cid:38) − W Hα values, itis unclear if the necessary parameters are physically realistic. We believe facular coverage fractions of ∼ ∼
5% during solar active periods (Shapiro et al. 2015) and plage coverage can reach ∼
8% (Foukal 1998). On the other hand, Shapiro et al. (2015) (see also Foukal 1998; Lockwood et al. 2007) findthat photometric variations for more active stars are spot-dominated and that the ratio of facular/plage coverage tospot coverage decreases with increasing activity level. Lanza et al. (2011) find that RV modulations in HD 189733’sspectrum are best modeled with values of Q = A fac /A spot ∼
0, i.e., they find no evidence of facular/plage effects onthe measured RV values. Furthermore, Lanza et al. (2011) find spot filling factors of ∼
1% are able to reproduce HD189733’s photometric variations across ∼ Q = 0 . ∼
8% facular/plage coverage.The larger question for H α contrast models is what constitutes a reasonable value of q fac . We have shown that q fac (cid:38) . W Hα similar to what is observed. If this is the case, then the normalor quiescent facular/plage regions on HD 189733 have H α emission line strengths similar to moderate flaring regionson the Sun. Magnetic fields play an important role in heating the chromosphere and producing bright regions in theatmosphere (e.g., Hansteen et al. 2007). Indeed, simulations of H α brightness in solar active regions show that thebrightest regions correspond to the strongest local magnetic field strengths (Leenaarts et al. 2012). HD 189733 has amuch stronger global magnetic field than the sun, with radial field values reaching 30-40 G across much of the stellarsurface (Fares et al. 2010). Thus the larger field values, and consequently heating rates, could have a significant effecton the emission strength of H α in active regions.A final consideration is the requirement that the facular/plage regions be concentrated very close to the planetarytransit chord, or a latitude very near +40 ◦ . Active latitudes are known features of the solar surface (e.g., Vecchio etal. 2012) and have been inferred via photometric transits for other stars as well (Sanchis-Ojeda & Winn 2011; Kirk etal. 2016). Lanza et al. (2011) find that spots near ≈ ◦ − ◦ are able to best reproduce the measured RV variations.Thus it is not unreasonable that the facular/plage regions are located near the transit chord. The specificity required,however, to reproduce any of the observed W Hα transits is concerning, especially if active latitudes migrate as afunction of time.Although we cannot conclusively distinguish between models in which the observed W Hα transits are due to absorp-tion by the planet or from the contrast effect, we believe the parameter values required to reproduce the observationsare rather specific and likely do not represent the average stellar disk of HD 189733. Furthermore, the contrast effectcannot explain the W Hα values seen in absorption immediately before the 2013 July 4 transit, before and after the2015 August 4 transit, and immediately after the 2006 August 21 transit, i.e., these extended transits are sugges-tive of absorbing circumplanetary material. In addition, we demonstrated in Cauley et al. (2017) that these pre-and post-transit signatures are abnormal and must be related to the planetary transit, further making the case fora circumplanetary origin. More detailed modeling of H α spectral line profiles in faculae and plages for active starsare needed to determine if the large core strengths used here are realistic. For now, however, we favor the planetaryinterpretation. H α VELOCITY MEASUREMENTS AND MODELS OF PLANETARY ROTATIONBarnes et al. (2016) presented an analysis of the same three HARPS transits presented in this work. They note atrend in the velocity centroids of the H α transmission spectra, when calculated in the frame of the planet, movingfrom red-shifted to blue-shifted (their Figure 3). This is cited as evidence that the absorption may not arise in theplanetary atmosphere since the absorption line profiles should be centered at zero velocity in the planetary referenceframe. In this section we present measurements of the H α transmission spectrum velocity centroids and models ofatmospheric absorption which include planetary rotation. We demonstrate that velocities in the upper atmosphere ofHD 189733 b might explain the in-transit velocity trends identified by Barnes et al. (2016).To date, only a handful of studies have presented observational signatures of atmospheric dynamics for massiveexoplanets. The first detection of a day-to-night side wind, a result consistent with predictions by a variety of detailedhot Jupiter circulation models (e.g., Showman & Guillot 2002; Showman et al. 2009; Rauscher & Menou 2010; Menou& Rauscher 2010; Rauscher & Kempton 2014), was made by Snellen et al. (2010) for HD 209458 b. They observed4a ∼ − blue-shifted offset in a CO absorption signal which matches closely with the magnitude of the flowvelocity from atmospheric models (e.g., Showman et al. 2008). Snellen et al. (2014) measured an equatorial rotationalvelocity of 25 km s − for the young massive planet β Pic b using cross-correlated thermal signatures of CO and H O.Wyttenbach et al. (2015) measured an 8 ± − blue-shift in the Na I doublet and suggested that the large velocitymight be the result of high-altitude winds.Most recently, studies by Louden & Wheatley (2015) and Brogi et al. (2016) have demonstrated detections of theplanetary rotational velocity and day-to-night side winds of HD 189733 b. For Na I , Louden & Wheatley (2015) alsoclaim a detection of a spatially resolved, eastward super-rotating equatorial jet. Brogi et al. (2016) search for a similarfeature in their near-IR CO measurement but are unable to place constraints on a jet velocity. Both studies findplanetary rotational velocities that are consistent with synchronous rotation and day-to-night side wind speeds of ∼ − . The results of Brogi et al. (2016) and Louden & Wheatley (2015) demonstrate the exciting possibility ofmeasuring the atmospheric dynamics of hot Jupiter atmospheres using ground-based, high-resolution spectra.In Cauley et al. (2016) we presented an H α velocity measurement (see Equation 7 below) and the correspondingin-transit v Hα values for two Keck HIRES transits. Barnes et al. (2016) investigated the H α line velocities of the threeHARPS data sets examined in this paper. They present velocity profiles as a function of orbital phase (see Figure 3 ofBarnes et al. 2016) but do not calculate velocities for individual absorption line profiles. They reference visual featuresin the velocity maps as evidence of trends in the line velocities across the transit. Here we present explicit velocitymeasurements of the individual HARPS and Keck spectra. As we show below, the individual HARPS transmissionspectra are fairly noisy and result in highly uncertain values of v Hα . The Keck velocities are more tightly constraineddue to the much higher signal-to-noise of the H α spectra.We perform all of our velocity analysis in the stellar rest frame. After correcting the observed spectra for the systemradial velocity and the Earth’s heliocentric motion, this is also the frame of the observer. Shifting the transmissionspectra by the planetary radial velocity can confuse and mask atmospheric velocities that may be contributing to theline profile, as we demonstrate below.The line velocity index from Cauley et al. (2016) is defined as: v Hα = +40 (cid:80) v = − v (1 − F v ) (cid:80) v = − (1 − F v ) (7)where F v is the transmission spectrum flux at velocity v . The index is essentially the velocity vector v weighted by thesquare of the transmission spectrum flux. The square of the flux is chosen to provide stronger weight to deeper portionsof the transmission spectrum. We only calculate v Hα for observations that show ≥ σ absorption. The uncertainty onthe individual v Hα points is calculated by taking the standard deviation of the mean of velocities corresponding to the(1 − F v ) values that comprise 68% of the total weight. This produces larger uncertainties for lines where the weightsare comparable at many different velocities, which is the case for the noisy HARPS S T profiles. The measured v Hα values are shown in Figure 18. The typical HARPS uncertainties are ∼ − while the Keck uncertainties are ∼ − , although they are slightly larger for the 2006 August 21 transit.It is clear from Figure 18 that there is no obvious v Hα trend across epochs, even for the transits that show consistent W Hα values in absorption (e.g., 2007 July 20, 2013 July 4, and 2015 August 4). The only date that shows any transitpattern is 2015 August 4: the velocities change from slightly blue-shifted to slightly red-shifted from the first halfof the transit to the second, although the mean of v Hα for each half of the transit is only different from zero at the ∼ − σ level. The 2013 July 4 transit shows a mild blue-shifted offset of ∼ − but v Hα values from the firsthalf of the transit are plagued by weak W Hα values. Little information is gleaned from the partial transits. We willexplore the 2015 August 4 trend using models of planetary rotation and velocities as a product of the contrast effectfrom Section 6. 7.1. Planetary rotation models
In order to explore physical scenarios for the H α line velocities presented above, we have simulated transmissionspectra through a rotating planetary atmosphere. We use the same stellar and planetary parameters that werepresented in Section 6 and the same PHOENIX model spectra are used as the intrinsic stellar spectra. We neglectactive regions on the stellar surface and assume a pure photosphere (see Section 7.2). The stellar radial velocity as afunction of in-transit time is included in the line profile calculations.5 −60 −40 −20 0 20 40 60t−t mid (minutes)−7.00.007.00 v H α ( k m s − ) −7.00.007.00 v H α ( k m s − ) −7.00.007.00 v H α ( k m s − ) −7.00.007.00 v H α ( k m s − ) −7.00.007.00 v H α ( k m s − ) −7.00.007.00 v H α ( k m s − ) −7.00.007.00 v H α ( k m s − ) Figure 18 . Values of v Hα from Equation 7 for all in-transit points showing 1 σ absorption. The uncertainties for most of the HARPSmeasurements are ∼ − while the Keck uncertainties are ∼ − . The vertical dashed lines mark the transit contact point.There are no clear patterns in the velocity measurements for any of the transits except 2015 August 4. The planetary atmosphere is assumed to be of uniform density. This choice is motivated by the models of Christieet al. (2013) who found that the number density of n = 2 hydrogen was approximately constant across 3-4 orders ofmagnitude in atmospheric pressure. Assuming a hydrostatic versus uniform density atmosphere has little effect on thesimulations. The planetary H α absorption profile is approximated as a velocity-broadened delta function where thebroadening parameter is labeled b (Draine 2011; Cauley et al. 2015, 2016). The planet is assumed to rotate rigidlythroughout the atmosphere.We have also investigated the effects of superrotating equatorial jets (Showman et al. 2008; Rauscher & Menou 2010)and find that they produce similar effects as rigid rotation, although the jet velocities required to produce the sameline profiles are larger than the rotational velocities since the jets occupy a smaller portion of the atmosphere. We donot include jets explicitly in the rest of the model discussion but we will discuss them as part of the general effect of6large rotational velocities. −6 −4 −2 0 2 4 6x (R p ) t−t mid =−30 minutes −6−4−20246 z ( R p ) RV=−8.8 km s −1 v H α =−2.6 km s −1 −40−20 0 20 40Velocity (km s −1 ) −6 −4 −2 0 2 4 6x (R p ) t−t mid =0 minutes RV=0.0 km s −1 v H α =0.1 km s −1 −40−20 0 20 40Velocity (km s −1 ) −6 −4 −2 0 2 4 6x (R p ) t−t mid =30 minutes RV=8.8 km s −1 v H α =2.7 km s −1 −40−20 0 20 40Velocity (km s −1 ) Figure 19 . Example of how planetary rotation affects the measured transmission spectrum. The colors are representative of the radialvelocity of the colored portion. Darker colors represent larger velocities. The stellar rotational velocities are weighted by the limb darkenedintensity to visually represent the weighted velocity contribution of the stellar disk. The bulk motion of the planetary disk is given a singlecolor and is given in the upper-right of each panel. The measured v Hα from Equation 7 for the inlaid magenta transmission spectrum isalso given in the upper-right corner. The v Hα values are much lower than the planetary RV for times when the planet is near the limb,suggesting that any planetary rotation will reduce the measured line velocities. This example shows absorption lines for an atmospherewith T exo = 7800, v rot = 10 km s − , b = 4 . − , and ρ = 3 . × − g cm − . Note that the individual v Hα values do not changemuch for different values of T exo , b , and ρ ; it is dominated by v rot . Figure 19 shows the effect of a rotating atmosphere on the transmission spectrum for a prograde planetary orbit.Upon ingress (left panel) the portion of the planet’s atmosphere dominating the transmission spectrum is moving away from the observer while the planet’s bulk motion is towards the observer. The net effect is to produce ameasured centroid velocity, or in our case v Hα , that is significantly less than the bulk planetary velocity. This wasfirst demonstrated by Miller-Ricci Kempton & Rauscher (2012) (their Figure 8) and reiterated in Louden & Wheatley(2015) (their Figure 2). The same effect is seen upon egress (right panel).Simulated v Hα curves for the entire transit are shown in Figure 20 where the effect of higher rotational velocitiescan be seen in the suppression of v Hα relative to the line-of-sight orbital velocity upon ingress and egress. For thehighest rotational velocities, the atmosphere is moving fast enough to cause v Hα to have the opposite sign comparedto the planet’s bulk motion. We note that for strong atomic lines even the case of no rotation produces suppressedline velocities due to the CLVs discussed in Section 6. For transits that are sampled asymmetrically in time, CLVscould produce velocity shifts in the average transmission spectrum (e.g., see Louden & Wheatley 2015; Wyttenbachet al. 2015).Figure 20 shows that for sufficiently large v rot the measured v Hα values can be very small or even have the oppositesign relative to the planetary bulk velocity. While these transit curves cannot explain the erratic velocities of theHARPS data or the 2006 August 21 Keck data, they may provide evidence for what we are seeing in the 2013 July 4and 2015 August 4 Keck transits and perhaps the first half of the 2013 June 3 transit.Figure 21 shows three different rotation models plotted against the 2015 August 4 v Hα values. Rotational velocities (cid:46) − do not match the data well; velocities between 10-12 km s − are required to produce the small v Hα valuesbetween first and second and third and fourth contact. Values of v rot (cid:38)
12 km s − begin to produce v Hα values thatare too small, i.e., of the wrong sign. The models do not do a good job of reproducing v Hα near −
20 and +20 minutes;all of the models produce v Hα larger than the measured values. Overall, however, these models demonstrate that largevelocities in the extended atmosphere can produce the small observed velocities.7.2. Velocities from contrast models
It is possible for certain contrast model scenarios to produce v Hα values similar to what is observed for the 2013July 4 and 2015 August 4 transits. We have calculated v Hα for the contrast model line profiles. The v Hα values forthe contrast models from Figure 16 (left panel), Figure 14 (middle panel), and Figure 15 (right panel) are shown in7 −60 −40 −20 0 20 40 60t−t mid (minutes)−10−50510 v H α ( k m s − ) rot (km s −1 ) −60 −40 −20 0 20 40 60t−t mid (minutes)−10−50510 v H α ( k m s − ) Figure 20 . H α transmission spectrum velocity measurements, defined by Equation 7, as a function of time from mid-transit. The colorindicates the equatorial planetary rotational velocity, the values of which are specified in the inset color bar. The transit contact pointsare marked with vertical dashed lines. As the planet rotates faster, absorption lines in the planetary atmosphere are shifted further awayfrom the planetary orbital velocity upon ingress and egress, creating the slope changes in v Hα seen near ±
25 minutes. Even the case of norotation shows a depressed line velocity compared with the orbital velocity. This is due to CLVs described in Section 6. This set of modelswas calculated with the same atmospheric parameters used in Figure 19.
Figure 22. The contrast model line profiles show trends similar to the 2015 August 4 data, although only the veryactive case (middle panel) shows the blue-shifted to red-shifted pattern from the first half of the transit to the secondhalf. Although not shown here, all of the other cases explored in Section 6 show something similar to the left panel ofFigure 22: large red-shifted velocities during the first half of the transit and blue-shifted velocities during the secondhalf. More active cases than the middle panel tend to move the other direction: larger and larger blue-shifted velocitiesand then red-shifted velocities during the second half.Overall, the velocities from the contrast models do not reproduce the observed 2015 August 4 velocities exceptfor the case of a very active stellar surface with the active regions centered very near the planet’s transit chord (seeFigure 14). For this reason, and those given concerning the planetary rotation models, we do not believe the contrastvelocities represent a convincing explanation for the observations. However, further in-transit observations should beconducted to strengthen or reject this argument.7.3.
Discussion of the rotation models and H α velocities The rotation models presented here are meant to demonstrate that the low velocities, i.e., much less than the in-transit planetary line-of-sight velocity, observed in the H α transits are not necessarily intrinsic to the star because oftheir magnitude. While very hot Jupiters such as HD 189733 b are generally assumed to be synchronously rotating,there are no measurements of hot Jupiter rotation rates outside of the studies done by Brogi et al. (2016) and Louden &Wheatley (2015). Both Brogi et al. (2016) and Louden & Wheatley (2015) found evidence for a synchronously rotatingHD 189733 b. However, these measurements were made in CO (Brogi et al. 2016) and Na I (Louden & Wheatley2015). The CO measurements probe much higher pressures deeper in the atmosphere than H α and Na I , which probepressures of 10 − -10 − bar (Christie et al. 2013; Wyttenbach et al. 2015). The Na I measurements by Louden &Wheatley (2015) do not examine the velocities of individual observations and instead fit the average in-transit lineprofile. Louden & Wheatley (2015) also do not include differential limb darkening, opting to use a broadband limb Brogi et al. (2016) measure the strongest signal in CO but also marginally detect H O. −60 −40 −20 0 20 40 60t−t mid (minutes)−4−2024 v H α ( k m s − ) v rot =6 km s −1 v rot =9 km s −1 v rot =12 km s −1 Figure 21 . Comparison of planetary rotation models with the observed v Hα values from the 2015 August 4 transit. Three representativerotation models from Figure 20 are shown and the plot scale is reduced to show the detailed shape of the model curves. The modelsproduce essentially identical v Hα values between second and third contact and do a poor job of matching the data for times immediatelyafter second contact and immediately before third contact. However, the v rot =9 and 12 km s − models show similar v Hα values comparedwith the data between first and second and third and fourth contacts. −60 −40 −20 0 20 40 60t−t mid (minutes)−505 v H α ( k m s − ) Contrast model2015−08−04A sp =0.02, Q=0.4, r fac =4.0, θ =40 o −60 −40 −20 0 20 40 60t−t mid (minutes) A sp =0.03, Q=0.15, r fac =4.0, θ =40 o −60 −40 −20 0 20 40 60t−t mid (minutes) A sp =0.03, Q=0.15, r fac =4.0, θ =30 o Figure 22 . Velocities from the contrast models (blue circles) overplotted on the v Hα values from the 2015 August 4 transit. The left panelcorresponds to the contrast model in Figure 16, the middle panel to Figure 14, and the right panel to Figure 15. The contrast velocitiesfrom the middle panel do a fair job of reproducing the observed v Hα values. Other contrast scenarios besides the very active case do notreproduce the observations well and in fact show the opposite trend, i.e., a transition from red-shifted to blue-shifted velocities from thefirst half to the second half of the transit. ∼ − in the upperatmospheres of both a synchronously and a quickly rotating HD 189733 b. Their models were calculated at pressures of10 -10 − bar, which begins to probe the potential H α formation region. Thus it is not implausible that even strongerwinds may form at higher altitudes or that unexplored atmospheric dynamics are contributing to the H α transmissionspectra.While the most active contrast model is able to produce v Hα values similar to what is observed for the 2015 August 4transit, we do not believe this is the correct model due to required specificity of the model parameters. In other words,matching the velocities requires all of the most active contrast model parameters to be accurate whereas matching onlythe W Hα observations provides more leeway in the active region coverage fraction and facular emission line strength.On the other hand, the 2015 August 4 transit does seem to be unique in that it shows the largest W Hα signal andthe strongest out-of-transit Ca II emission. Although it seems unlikely, we cannot definitively rule out the most activecase as an explanation for the both W Hα and v Hα .We emphasize that the H α transits do not show a consistent velocity signal across epochs and as a result no firmconclusions can be reached concerning their origin. The individual HARPS spectra are especially noisy and the velocityuncertainties derived here are large making them even less useful for understanding the in-transit signal. Furthershort-cadence H α transit observations are needed to work towards clarifying the velocity signal and, by extension, thein-transit H α absorption line profiles. SUMMARY AND CONCLUSIONSWe have presented an analysis of H α transmission spectra for 7 transits of HD 189733 b that span almost a decade.Five of these H α transits are from archival HARPS and Keck data, while two were previously analyzed by our group inCauley et al. (2015, 2016). Four of these transits show significant and consistent H α spectra in absorption throughoutthe transit while three others show strong variations that may be due to changes in stellar activity. The irregularchanges may also be due to transiting gas not bound to the planet. Our main conclusions are the following: • We do not find evidence of a clear relationship between the stellar activity level and the strength of the in-transitH α signal. The outlier in this case is the data from 2015 August 4 which shows both the strongest H α absorptionand the highest stellar activity level. If the absorption signal arises in the planetary atmosphere, this reveals apotential stellar activity threshold below which the ionizing flux from the star is not high enough to create largeH α transit depths. This should be investigated further with future transits and simultaneous UV measurementsof the stellar activity level. • We explored detailed simulations of the H α contrast effect for HD 189733 b. We find that large facular/plagecoverage fractions of (cid:38) −
10% and very strong facular/plage core emission strengths of q fac ∼ α observations. Furthermore, these facular/plage regions must be concentrated veryclose to the transit chord of the planet. Spots and filaments do not have a significant effect on the H α contrastspectrum. • Due to the specificity of the contrast parameters required to reproduce the measured W Hα values, combinedwith the natural explanation of absorption in the thermosphere (Christie et al. 2013), we favor a planetaryatmospheric origin for the H α transmission spectra. A similar argument can be made for interpreting the v Hα measurements. However, detailed models of active regions on active stars such as HD 189733 are needed tounderstand if the necessary contrast parameters are reasonable. • We have also explored the velocity centroids of the measured H α transmission spectra using models of planetaryrotation. We find that planetary rotational velocities of ∼ − are able to produce in-transit v Hα valuessimilar to those from the 2015 August 4 transit. These large velocities are representative of dynamics in thevery extended atmosphere and do not necessarily suggest that the planet is rotating faster than the synchronousrate. These models demonstrate that large atmospheric velocities can produce the small observed v Hα valuesand that they do not need to originate on the stellar surface. However, there is no consistent velocity patternacross epochs so the results of the rotation models cannot be broadly applied. Further short-cadence transits atvery high S/N are needed to test these ideas.0Future and current planet hunting missions, such as the Transiting Exoplanet Survey Satellite ( TESS ) and theground-based Kilodegree Extremely Little Telescope (KELT Pepper et al. 2007) and the Multi-site All-Sky CAmeRA(MASCARA Talens et al. 2017), have and will detect many hot planets transiting relatively bright stars. This willsignificantly increase the number of objects for which short-cadence, high-resolution optical transmission spectroscopycan be performed with 4-meter or 10-meter class telescopes. In addition, thirty-meter class telescopes will be able toperform similar observations on much fainter systems. Stars with high activity levels should be targeted as comparisoncases with HD 189733 b. Indeed, active stars may be the only systems with hot planets exhibiting H α absorptionin their extended atmospheres (Christie et al. 2013). These systems will also act as testbeds for disentangling thecontrast effect from true planetary absorption.Ground based H α observations of the extended atmospheres of hot planets offers a complimentary alternative tothe exospheric Hubble Space Telescope ( HST ) Ly α observations (Vidal-Madjar et al. 2003; Lecavelier des Etangs etal. 2010, 2012; Bourrier et al. 2013; Vidal-Madjar et al. 2013; Kulow et al. 2014; Ehrenreich et al. 2015). While theH α observations do not probe the escaping exosphere, this ground-based approach may become the best option formeasuring the base of evaporative flows due to the impending loss of HST and its spectroscopic UV capabilities. Itis also critical to develop ground based transmission spectrum programs in order to plan for and complement futurespace missions, such as the
Large UV/Optical/IR Surveyor ( LUVOIR ). HD 189733 is a benchmark for testing theusefulness of the H α diagnostic and the relationship between stellar activity and the planetary thermosphere. It isthus important to further investigate the observed HD 189733 signals, along with any future detections, to determinethe origin of the H α signal and strengthen or repudiate the arguments presented here. Acknowledgments:
We thank the referee for their critique of this manuscript, which helped improve the clarityand style. A portion of the data presented herein were obtained at the W.M. Keck Observatory from telescope timeallocated to the National Aeronautics and Space Administration through the agency’s scientific partnership with theCalifornia Institute of Technology and the University of California. This work was supported by a NASA Keck PIData Award, administered by the NASA Exoplanet Science Institute. The Observatory was made possible by thegenerous financial support of the W.M. Keck Foundation. The authors wish to recognize and acknowledge the verysignificant cultural role and reverence that the summit of Mauna Kea has always had within the indigenous Hawaiiancommunity. We are most fortunate to have the opportunity to conduct observations from this mountain. A. G. J. issupported by NASA Exoplanet Research Program grant 14-XRP14 2- 0090 to the University of Nebraska-Kearney.This work was completed with support from the National Science Foundation through Astronomy and AstrophysicsResearch Grant AST-1313268 (PI: S.R.). This work has made use of NASA’s Astrophysics Data System.REFERENCES
Ahn, K., Chae, J., Cho, K.-S., et al. 2014, SoPh, 289, 4117Aigrain, S., Pont, F., & Zucker, S. 2012, MNRAS, 419, 3147Andersen, J. M., & Korhonen, H. 2015, MNRAS, 3053, 3069Astudillo-Defru, N., & Rojo, P. 2013, A&A, 557, 56Barnes, J. R., Haswell, C. A., Staab, D., & Anglada-Escud´e, G.2016, MNRAS, 462, 1012Ben-Jaffel, L., & Ballester, G. E. 2013, A&A, 553, A52Berdyugina, S. V. 2005, Liv. Rev. Sol. Phys., 2, 8Berta, Z. K., Charbonneau, D., Bean, J., et al. 2011, ApJ, 736,12Boisse, I., Moutou, C., Vidal-Madjar, A., et al. 2009, A&A, 495,959Bouchy, F., Udry, S., Mayor, M., et al. 2005, A&A, 444, L15Bourrier, V., Lecavelier des Etangs, A., Dupuy, H., et al. 2013,A&A, 551, A63Boyajian, T., von Braun, K., Feinden, G. A., et al. 2015,MNRAS, 447, 846Brogi, M., de Kok, R. J., Albrecht, S., et al. 2016, ApJ, 817, 106Cauley, P. W., Redfield, S., Jensen, A. G., et al. 2015, ApJ, 810,13Cauley, P. W., Redfield, S., Jensen, A. G., & Barman, T. 2016,AJ, 152, 20Cauley, P. W., Redfield, S., & Jensen, A. G. 2016, AJ, acceptedChristie, D., Arras, P., & Li, Z.-Y. 2013, ApJ, 772, 144 Collier Cameron, A., Bruce, V. A., Miller, G. R. M., Triaud, A.H. M. J., & Queloz, D. 2010, MNRAS, 403, 151Cuntz, M., Saar, S. H., & Musielak, Z. E. 2000, ApJ, 533, L151Czesla, S., Klocov´a, T., Khalafinejad, S., Wolter, U., & Schmitt,J. H. M. M. 2015, A&A, 582, A51Dekker, H., D’Odorico, S., Kaufer, A., Delabre, B., &Kotzlowski, H. 2000, SPIE, 4008, 534Deng, N., Tritschler, A., Jing, J., et al. 2013, ApJ, 769, 112Di Gloria, E., Snellen, I. A. G., & Albrecht, S. 2015, A&A, 580,A84Draine, B. T. 2011,
Physics of the Interstellar and IntergalacticMedium (Princeton University Press, Princeton, NJ)Dumusque, X., Boisse, I., & Santos, N. C. 2014, ApJ, 796, 132Eker, Z., Brandt, P. N., Hanslmeier, A., et al. 2003, A&A,404,1107Ehrenreich, D., Bourrier, V., Bonfils, X., et al. 2012, A&A, 547,18Ehrenreich, D., Bourrier, V., Wheatley, P. J., et al. 2015, Nature,522, 459Fares, R., Donati, J.-F., Moutou, C., et al. 2010, MNRAS, 406,409Fossati, L., Haswell, C. A., Froning, C. S., et al. 2010, ApJ, 714,L222Fossati, L., Ingrassia, S., & Lanza1