Kepler sheds new and unprecedented light on the variability of a blue supergiant: gravity waves in the O9.5Iab star HD 188209
C. Aerts, S. Simon-Diaz, S. Bloemen, J. Debosscher, P.I. Papics, S. Bryson, M. Still, E. Moravveji, M.H. Williamson, F. Grundahl, M. Fredslund Andersen, V. Antoci, P.L. Palle, J. Christensen-Dalsgaard, T.M. Rogers
AAstronomy & Astrophysics manuscript no. hd188209-astroph © ESO 2017March 7, 2017
Kepler sheds new and unprecedented light on the variability of ablue supergiant: gravity waves in the O9.5Iab star HD 188209 (cid:63)
C. Aerts , , S. Símon-Díaz , , S. Bloemen , , J. Debosscher , P. I. Pápics , S. Bryson , M. Still , , E. Moravveji , M.H. Williamson , F. Grundahl , M. Fredslund Andersen , V. Antoci , P. L. Pallé , , J. Christensen-Dalsgaard , and T. M.Rogers , Instituut voor Sterrenkunde, KU Leuven, Celestijnenlaan 200D, 3001 Leuven, Belgiume-mail:
[email protected] Department of Astrophysics / IMAPP, Radboud University Nijmegen, 6500 GL Nijmegen, The Netherlands Instituto de Astrofísica de Canarias, 38200, La Laguna, Tenerife, Spain Departamento de Astrofísica, Universidad de La Laguna, 38205, La Laguna, Tenerife, Spain NASA Ames Research Center, Mo ff ett Field, CA 94095, USA Bay Area Environmental Research Institute, 560 Third Street W., Sonoma, CA 95476, USA Center of Excellence in Information Systems, Tennessee State University, 3500 John A. Merritt Blvd., Box 9501, Nashville, TN37209, USA Stellar Astrophysics Centre, Department of Physics and Astronomy, Aarhus University, DK-8000 Aarhus C, Denmark Department of Mathematics and Statistics, Newcastle University, UK Planetary Science Institute, Tucson, AZ 85721, USAReceived ; Accepted
ABSTRACT
Stellar evolution models are most uncertain for evolved massive stars. Asteroseismology based on high-precision uninterrupted spacephotometry has become a new way to test the outcome of stellar evolution theory and was recently applied to a multitude of stars,but not yet to massive evolved supergiants.Our aim is to detect, analyse and interpret the photospheric and wind variability of theO9.5 Iab star HD 188209 from
Kepler space photometry and long-term high-resolution spectroscopy. We used
Kepler scattered-lightphotometry obtained by the nominal mission during 1460 d to deduce the photometric variability of this O-type supergiant. In addition,we assembled and analysed high-resolution high signal-to-noise spectroscopy taken with four spectrographs during some 1800 d tointerpret the temporal spectroscopic variability of the star. The variability of this blue supergiant derived from the scattered-light spacephotometry is in full in agreement with the one found in the ground-based spectroscopy. We find significant low-frequency variabilitythat is consistently detected in all spectral lines of HD 188209. The photospheric variability propagates into the wind, where it hassimilar frequencies but slightly higher amplitudes. The morphology of the frequency spectra derived from the long-term photometryand spectroscopy points towards a spectrum of travelling waves with frequency values in the range expected for an evolved O-typestar. Convectively-driven internal gravity waves excited in the stellar interior o ff er the most plausible explanation of the detectedvariability. Key words.
Line: profiles – Techniques: spectroscopic – Techniques: photometric – Stars: massive – Stars: oscillations (includingpulsations) – Waves
1. Introduction
Stars born with su ffi ciently high mass to explode as supernovaat the end of their life have major impact on the dynamical andchemical evolution of galaxies. Appropriate models of such pre-supernovae are thus highly relevant for astrophysics. Unfortu-nately, the theory of their evolution is a lot less well established (cid:63) Based on photometric observations made with the NASA
Kepler satellite and on spectroscopic observations made with four telescopes:the Nordic Optical Telescope operated by NOTSA and the MercatorTelescope operated by the Flemish Community, both at the Observato-rio del Roque de los Muchachos (La Palma, Spain) of the Instituto deAstrofísica de Canarias, the T13 2.0m Automatic Spectroscopic Tele-scope (AST) operated by Tennessee State University at the FairbornObservatory, and the Hertzsprung SONG telescope operated on theSpanish Observatorio del Teide on the island of Tenerife by the Aarhusand Copenhagen Universities and by the Instituto de Astrofísica de Ca-narias, Spain than the one of low-mass stars that die as white dwarf. Di ff er-ences in the predictions of massive star evolution from variousmodern stellar evolution codes even occur already well beforethe end of the main-sequence (MS) phase (e.g., Martins & Pala-cios 2013).Despite the immense progress in the asteroseismic tuningof stellar models of various types of stars from high-precisionuninterrupted space photometry in the past decade (e.g., Chap-lin & Miglio 2013; Charpinet et al. 2014; Aerts 2015; Hekker& Christensen-Dalsgaard 2016, for reviews), we still lack suit-able data to achieve this stage for massive O-type stars andtheir evolved descendants, the B supergiants. Indeed, while theMOST and CoRoT missions did observe a few B supergiantsfor weeks to months (e.g., Saio et al. 2006; Aerts et al. 2010;Moravveji et al. 2012; Aerts et al. 2013), their pulsational fre-quencies were not measured with su ffi cient precision and / or theangular wavenumbers ( (cid:96), m ) of their oscillation modes could not Article number, page 1 of 14 a r X i v : . [ a s t r o - ph . S R ] M a r & A proofs: manuscript no. hd188209-astroph be identified (e.g., Aerts et al. 2010, for a detailed description ofmode identification in asteroseismology). A similar situation oc-curs for the earlier phases, given the absence of suitable highlysampled space photometric data with su ffi ciently long time basefor O stars, despite appreciable e ff orts (see, e.g., Buysschaertet al. 2015, for an updated summary). Hence, improvement ofthe input physics adopted in stellar models representing vari-ous evolutionary phases of massive stars is still beyond reach,in contrast to such achievements for low-mass stars (e.g., Bed-ding et al. 2011; Bischo ff -Kim & Østensen 2011; Foster et al.2015; Deheuvels et al. 2015, to mention just a few studies).There are several reasons why asteroseismology of evolvedstars in the mass range of supernova-progenitors, i.e., with birthmasses above some 8 M (cid:12) , is so di ffi cult to achieve. First andforemost, such stars have large radii and connected with this,their oscillation periods are several to tens of days. Any multi-periodic nonradial mode beating pattern therefore reaches peri-ods of years and such a time base is beyond the capacity of theMOST, CoRoT, and K2 space missions. This is also the reasonwhy gravity-mode pulsators in the core-hydrogen burning phase,which have periods of the order of half to a few days, couldonly be fully exploited seismically thanks to the nominal Ke-pler mission. Indeed, although their period-spacing pattern wasfirst discovered from 150 d of uninterrupted CoRoT data (Deg-roote et al. 2010), seismic modelling of MS gravity-mode pul-sators required at least a year of uninterrupted space photometry.Four years of
Kepler data of B and F stars led to interior struc-ture properties that cannot be explained with standard models, interms of interior rotation and mixing (e.g., Kurtz et al. 2014; Saioet al. 2015; Moravveji et al. 2015; Triana et al. 2015; Moravvejiet al. 2016; Murphy et al. 2016; Van Reeth et al. 2016; Schmid &Aerts 2016). It is thus to be expected that models of their evolvedcounterparts deviate even more from reality.A second reason that hampers asteroseismology of O starsand B supergiants is the fact that their variability is not onlycaused by heat-driven coherent stellar oscillations, but as wellby a time-variable radiation-driven stellar wind, rotational ef-fects, macroturbulence, and for very few of the youngest Ostars magnetic activity as well (Fossati et al. 2016). All thesephysical phenomena interact, often non-linearly and in a non-adiabatic regime. This leads to complex overall variability witheven longer time bases than a classical multiperiodic oscillation.Here, we present a study of HD 188209 (O9 Iab), the onlymassive supergiant that was monitored with the nominal
Kepler mission during a total time base of four years and about equallylong in ground-based spectroscopy. We first introduce the knownproperties of our target star and then discuss the long-term mon-itoring in space photometry and in ground-based spectroscopy.We provide evidence for variability with an entire spectrum ofsperiods of the order of half to a few days in the independentdata sets.
2. The O9 Iab supergiant HD 188209
Given its visual magnitude of 5.63, HD 188209 was the sub-ject of various observational variability studies so far. Thesemainly focused on spectroscopy and were limited to only fewspectra gathered with low sampling rates. Early spectroscopictime-series assembled by Fullerton et al. (1996) and Israelianet al. (2000) revealed line-profile variability in the UV and op-tical parts of its spectrum, with various time scales of the orderof days. Additional studies by Markova et al. (2005), Fullertonet al. (2006), and Martins et al. (2015a), revealed variability in both the photosphere and in the stellar wind, with seemingly un-correlated quasi-periodicities occurring in those two regimes.The spectral type assigned to HD 188209 is O9.5Iab (Wal-born 1972; Sota et al. 2011). It is included in the list of stan-dard stars for spectral classification (e.g., Walborn & Fitzpatrick1990; Maíz Apellániz et al. 2015). The fundamental parame-ters, mass-loss rate, and level of macroturbulence from recentanalyses based on the non-LTE codes CMFGEN (Martins et al.2015a,b) and FASTWIND (Markova et al. 2005, Holgado et al.2017, in preparation), both including the e ff ects of line blan-keting and the stellar wind, are listed in Table 1. While bina-rity dominates the evolution of massive stars (Sana et al. 2012),HD 188209 was found to be a single star (Martins et al. 2015b).In line with the low incidence of surface magnetic fields in OBstars, HD 188209 led to a non-detection of such a field (Grunhutet al. 2017).Given its presence in the nominal Field-of-View (FoV) ofthe Kepler satellite, we have embarked on a unique long-termmonitoring study of HD 188209, based on space photometry andground-based high-resolution spectroscopy.
3. Scattered-light
Kepler photometry
The
Kepler
CCDs saturate for K p ∼ .
5, the precise value de-pending on the CCD and the target location on that CCD (Kochet al. 2010). However, it turned out that the saturated flux isconserved to a very high degree as long as su ffi ciently largemasks are placed around the star under study. This led to sev-eral variability studies of bright stars with unprecedented pho-tometric precision and duration of the light curves by means ofcustomized masks (e.g. Kolenberg et al. 2011; Metcalfe et al.2012; Tkachenko et al. 2014; Guzik et al. 2016). Meanwile, nu-merous bright stars are also studied with the refurbished Kepler mission K2, from co-adding carefully masked smear flux in theCCD rows of data taken with ultra-short exposure times (Whiteet al., submitted), a technique that was verified successfully onnominal
Kepler data (e.g., Pope et al. 2016).Here, we present a di ff erent use of the original Kepler satel-lite, following our idea to perform scattered-light aperture pho-tometry of HD 188209, despite the star not being on active sil-icon. With K p ∼ .
5, HD 188209 was indeed a nuissance forexoplanet hunting with the nominal
Kepler mission and it wastherefore placed carefully in between the active CCDs pointingto the nominal FoV. Nevertheless, it is so bright that its scatteredlight does “pollute” the CCDs (cf. Fig. 1). As only targets on ac-tive silicon are eligible for the
Kepler and K2 Guest Observerprograms, we had to proceed in a di ff erent way and requested“informally” from the Kepler
GO o ffi ce to place masks in thevicinity of HD 188209, with the reasoning that its variability be-haviour will be present and detectable in the scattered-light pho-tometry.For the current analysis, we used two apertures placed in thescattered light of the star. The two masks placed on active siliconcorrespond to the targets labelled KIC 10092932 (to the right)and KIC 10092945 (to the left) of HD 188209 in Fig. 1, hereaftercalled mask 32 and mask 45, respectively. Mask 45 contains afaint star of K p ∼ .
35, which is 7.85 magnitudes fainter thanHD 188209 itself; mask 32 does not contain any known sourceand is therefore a “pure” probe of the scattered light of the su-pergiant. It has K p = . Article number, page 2 of 14. Aerts et al.:
Kepler photometry and high-resolution spectroscopy of the O9.5 supergiant HD 188209
Table 1.
Spectroscopically derived properties of HD 188209 reported in the recent literature, based on a non-LTE analysis. Typical error bars are1 000 K for T e ff , 0.2 dex for log g , and several km s − for v sin i and v macro . Reference T e ff log g [N(He) / N(H)] log ˙
M v ∞ v sin i v macro (K) (cgs) (M (cid:12) yr − ) (km s − ) (km s − ) (km s − )Markova et al. (2005) 31 000 3.1 0.12 -5.78 1650 87 –Martins et al. (2015a,b) 29 800 3.2 – -6.40 2000 45 33Holgado et al. (in preparation) 31 100 3.0 0.14 – – 57 75 Fig. 1.
Position of the two apertures (45: left), 32 (right) in the
Kepler
FoV capturing the scattered light of HD 188209 (situated in the black bandabove the red cross in between active silicon).
Fig. 2.
An example of pixel masks used to create three versions of the light curves for star 45 (left) and 32 (right). The detected light curve isplotted within each pixel and represents an addition of 270 individual exposures of 6.54 s, added onboard of the satellite. In long-cadence mode,such pixel data are downloaded from the spacecraft in time stamps of 29.42 minutes. The value listed in every pixel, as well as the backgroundgrey scale, indicate the SNR of the flux in the pixel. The red dots in the pixels with green and blue borders were used to extract the standard
Kepler light curves. We also computed light curves based on the data in the yellow pixels alone and from adding the green, blue, and yellow pixels.
Masks of a selected
Kepler quarter showing the light curvesin the individual pixels as downloaded from the spacecraft foreach of the two apertures are shown in Fig. 2. These show thetypical time series traces with pixel-to-pixel variations as fornormal targetted stars (, e.g.,)Papics2014, except that the fluxlevels are lower here. This is particularly so for mask 32 giventhat it concerns scattered light rather than an actual target. Thesignal-to-noise ratio (SNR hereafter) of the detected flux in ev-ery pixel is indicated in each of the boxes. The pixels with darkgreen or blue borders with the red dots as flux values (stored ine − s − ) were used to extract the “standard” Kepler light curves oftargets as provided in MAST. We did not only use these lightcurves but also constructed two new ones that are based on theyellow pixels in which significant signal coming from the super-giant is present as well; SNR values above 100 typically corre- archive.stsci.edu/kepler/data − search/search.php spond to useful signal to be added for stellar variability studies.Adding the flux in those yellow pixels in addition to the standardMAST pixels leads to cleaner detrending and less contamina-tion by instrumental e ff ects at low frequencies (Tkachenko et al.2013; Pápics et al. 2014). For each of the two custom masks,we extracted two light curves in addition to the standard MASTcurves: one based on the yellow flux alone and another one basedon the green + blue + yellow flux. For each of the two masks, thevariability behaviour of these three light curves is the same interms of periodic behaviour, which is evidence that this variabil-ity is dominated by the scattered light of the supergiant and notby other sources in the pixel masks.In the rest of the paper, we proceed with the light curvescontaining maximal signal-to-noise ratio (SNR hereafter). Theseare the ones deduced from adding all the flux in the red, blue,and yellow pixels of masks 32 and 45 as shown in Fig. 2. Fig-ure 3 compares these two light curves, where mask 45 was Article number, page 3 of 14 & A proofs: manuscript no. hd188209-astroph
Fig. 3.
Light curves in masks 32 (grey circles) and 45 (black circles)covering the full length (upper), quarters 4 and 5 (middle) and a zoomof 20 d. observed during 1460 d from 13 / / / / / / / / Despite it being scat-tered light, this is by far the best light curve ever obtained for ablue supergiant.
It can be seen in the lower panel of Fig. 3 that
Fig. 4.
Amplitude spectrum for the
Kepler light curves of masks 32(grey) and 45 (black). The inset compares the spectrum of the
Kepler light curve from mask 32 (grey) with the
Hipparcos light curve (black)over twice the frequency range. The red dashed and blue dot-dot-dot-dashed lines indicate four times the average noise level computed overthe frequency range [0 ,
6] d − , while the red dot-dashed line in the insetrepresents four times the average noise level computed over the samerange [0 ,
6] d − of the Hipparcos data. the two curves are very similar in terms of temporal variability,but that the amplitude of the variability is di ff erent. This is asexpected given the di ff erence in flux level within the masks andthe fact that mask 45 contains a faint star while mask 32 doesnot.We computed the Fourier transform of the two light curvesshown in Fig. 3 to obtain the amplitude spectra up to the Nyquistfrequency. This resulted in a flat distribution of noise beyond6 d − and a clear amplitude excess above the noise level at lowfrequency for both light curves. A part of the spectrum is shownin Fig. 4, where the horizontal lines are situated at four times theaverage noise level computed over the range [0 ,
6] d − . Giventhat mask 32 is the “purer” one in terms of only scattered lightof HD 188209, its amplitude is dominant over the one obtainedfrom mask 45. However, both light curves are fully consistentwith each other in the sense that significant variability is de-tected over the whole frequency range [0 ,
2] d − in the form ofan excess of amplitude for numerous close frequencies. This is afrequency regime typical for gravity waves in massive stars.The amplitude spectrum in Fig. 4 is entirely di ff erent in na-ture from the one of the rotationally variable B5 supergiantHD 46769 observed with CoRoT during 23 d and revealing onlyone frequency and its harmonics corresponding to a rotation pe-riod of 4.8 ± . ff erent from those found for Kepler data ofB dwarfs pulsating in coherent heat-driven gravity modes (e.g.,Pápics et al. 2015, 2017) as well as from those of O dwarfs withvariability assigned to rotational modulation or heat-driven co-herent non-radial pulsations (see Table 3 in Buysschaert et al.2015, for an overview).
Article number, page 4 of 14. Aerts et al.:
Kepler photometry and high-resolution spectroscopy of the O9.5 supergiant HD 188209
Fig. 5.
Short-time Fourier transforms of the
Kepler light curves in mask45 (upper) and 32 (lower) for all observed quarters, for a 30-day timewindow and progressing in the time series with a step of 3 days, zoomedinto the region of low-frequency amplitude excess. Brighter colours rep-resent higher amplitudes.
Since we are dealing with a photometric curve deduced fromscattered light, a comparison with other photometric data of theactual star is meaningful. The only publicly available light curveof HD 188209 to make such a comparison is the one assembledby the Hipparcos satellite, which is of similar total length buthas only 113 data points spread over 1156 d, with far less pre-cision than the
Kepler scattered light data. This data set was already analysed by Israelian et al. (2000, see their Fig. 3) andrevealed variability but did not lead to clear periodicity. We re-computed the amplitude spectrum of the Hipparcos light curveand compared it with the
Kepler one of mask 32, which has sim-ilar length, in the inset of Fig. 4. Although the significance of theamplitude excess in the Hipparcos data is only marginal at best,low-frequency excess in [0 ,
2] d − is also hinted at in this inde-pendent photometric data set, at a level of some 14 mmag, whilewe found 3.6 mmag from the scattered light in mask 32.The photometric amplitude spectrum of the scattered light ofHD 188209 is remarkably similar to the one obtained from the Kepler data of the bright primary B0 III star (Lesh 1968) of theeclipsing binary V380 Cyg. V380 Cyg A’s variability behaviourwas interpreted in terms of excess power due to internal gravitywaves (see Fig. 3 in Tkachenko et al. 2014). This star has simi-lar log g as HD 188209 but is much less massive (some 12 M (cid:12) )and cooler (21,700 K). Similarly shaped excess power, althoughcovering a broader range in frequency, were also found for threeunevolved O stars hotter than HD 188209 (e ff ective temperaturesbetween 35,000 K and 43,000 K) observed by the CoRoT satel-lite (Blomme et al. 2011) and explained in terms of internal grav-ity waves caused by core convection in Aerts & Rogers (2015).Even though further analysis of the excitation layer of the waves– convective core or convection zone in the stellar envelope dueto the iron opacity bump – needs to be sought, variability dueto the stellar wind would results in few quasi-periodicities ona timescale of a day to a week (Kaper et al. 1996), while herewe are dealing with an entire spectrum of excited low frequen-cies without a dominant base frequency. For these four stars inthe literature, the nature of the variability was revealed throughshort-time Fourier transformations (STFTs; Fig. 3 in Tkachenkoet al. (2012) and Fig. 6 in Blomme et al. (2011) for these fourstars). Figure 5 shows STFTs of HD 188209 based on the two Kepler data sets. They were calculated using a 30-day time win-dow, each time progressing in the time series with a step of 3days, but the results were checked to remain the same for othervalues of the window and step. Just as for the four OB stars inthe literature, it is clear from Fig. 5 that the signal in these STFTsof HD 188209 is not due to multimode beating of stable, phase-coherent non-radial pulsation modes because the latter’s STFTslook quite di ff erent (see, e.g., Fig. 5 in Degroote et al. 2012).We conclude that the Kepler scattered light photometry ofHD 188209 points towards a fifth case of highly significant vari-ability with a multitude of low frequencies for a hot massive star;it is the first case of a massive supergiant for which such type ofvariability is revealed. All five stars in which this phenomenonhas been found in high-cadence space photometry are moder-ate rotators, having v sin i values between 50 and 100 km s − andrequire considerable macroturbulent broadening to explain theline-profile shapes in high-resolution spectroscopy. In order toproperly interpret the variability detected already early on in thescattered light of HD 188209, we initiated long-term ground-based follow-up high-resolution spectroscopy and included thestar in the IACOB project (Simón-Díaz & Herrero 2014).
4. Long-term high-resolution spectroscopy
HD 188209 was added to the large sample of OB-type starsacross the entire evolutionary path for long-term ground-based high signal-to-noise spectroscopic monitoring (Simón-Díaz et al. 2015, 2017). In contrast to the
Kepler photometry,such type of data have a low duty cycle and limited time cov-erage per night. This inevitably leads to heavily gapped spec-troscopic time series, where the interpretation of the variabil-
Article number, page 5 of 14 & A proofs: manuscript no. hd188209-astroph ity encounters the challenge to deal with daily alias structuresin the Fourier domain. Moreover, for a star as HD 188209, wewill be dealing with quasi-periodicities occurring due to a mix-ture of photospheric and wind variability. This was indeed al-ready revealed for HD 188209 by Martins et al. (2015a,b), whostudied it from a time-series of spectro-polarimetry consistingof 27 spectra gathered over a time span of 9 d. They consideredtwelve spectral lines (three Balmer lines, seven helium lines, andtwo metal lines) and found eight to be variable, including thosethat form (partially) in the wind, while four photospheric linesturned out to be stable in their data. Martins et al. (2015a) con-cluded that the photospheric line profiles of HD 188209 seem-ingly change on time scales from an hour to days, while its windvariability reveals longer-term periodicity (see also Markovaet al. 2005).No obvious connection between the quasi-periodicities in thephotosphere and in the wind was found for HD 188209 so far.However, all the data sets available in the literature have toofew spectra, too sparse sampling, and too limited time base tofind periodicities with appropriate precision. Given the limitedamount of spectra, all these studies only considered the variabil-ity inside the line profile without performing frequency analysis,by means of the so-called temporal variance spectrum (Fullertonet al. 1996). This quantity is very useful to estimate the overalllevel of variability that is present in a small series of spectrallines assembled with sparse sampling. Its capacity to unravel the cause of this variability is limited. Here, we shall be focusing onthe temporal line variability in a large data set of spectra with along time base, with the aim to investigate the physical reasonsof its origin. We do this by considering specialised line-profilequantities that are specifically designed to allow for time-seriesanalysis without being a ff ected by uncertainties due to spectrumnormalisation, as will be discussed below.We take a major step ahead compared to the spectroscopyresults in the literature from new extensive long-term spectro-scopic time series assembled with four instruments: – the fiber-fed HERMES échelle spectrograph attached to the1.2m Mercator telescope at Roque de los Muchachos, Is-land of La Palma, Spain, operated in the high-resolution(R = – the high-resolution FIbre-fed Échelle Spectrograph (FIES)attached to the Nordic Optical Telescope (NOT) at Roque delos Muchachos, Island of La Palma, Spain, operated in thehigh-resolution (R = – the T13 2.0m Automatic Spectroscopic Telescope (AST)with the fiber-fed cross-dispersed échelle spectrograph inthe resolution mode of R = – the prototype SONG node échelle spectrograph, operationalat Observatory del Teide on Tenerife, Spain (Grundahl et al.2007; Uytterhoeven et al. 2012; Grundahl et al. 2017). TheSONG observations of HD 188209 were obtained in ThArmode with slit 5 and have R =
77 000.A summary of the characteristics of all the data sets analysed inthis paper is given in Table 2.These four instruments were constructed with di ff erent pur-poses and are attached to telescopes of di ff erent sizes. The spe-cific aim of HERMES is to obtain maximal capacity to detecttime-resolved line-profile variability such that the SNR is im-portant. FIES is a similar spectrograph but delivers lower res-olution and is less well temperature stabilised. The main goal of SONG is to gain maximum precision in radial-velocity varia-tions by means of an iodine cell for solar-like oscillation studies(Grundahl et al. 2017), but it also o ff ers a ThAr option, whichwe have used here. As we will show below (see Figs 7 and 8) thelower resolving power of AST, combined with the lower SNR itdelivers, imply that these data are at the limit of what we needto properly detect and interpret the variability of HD 188209, butadding this data set gave compatible results so we kept them inthe analysis and adopted an approach taking into account theSNR. Indeed, the SNR levels achieved with the four instrumentsare di ff erent for similar integration times and provide us with away to properly weigh the data when intepreting the variability.In the following, we focus mainly on those spectral lines thatare available for the four independent spectroscopic data setslisted in Table 2 and that do not su ff er from order merging un-certainties in their line wings. This is the case for nine spectrallines. We also considered the Si iii γ as important diagnos-tic of the gravity behaviour of the star.For each spectral line, we computed its line moments as gooddiagnostics for variability by adopting their definition by Aertset al. (1992). It is crucial to define the optimal integration limitsin the line wings when computing the moments for the detec-tion and interpretation of low-amplitude line-profile variability(e.g., Chapters 5 and 6 in Aerts et al. (2010) and Zima (2008),for extensive descriptions and publicly available software). Sincewe aim to merge the moments of the same spectral lines ob-tained with four di ff erent instruments, we determined the mostoptimal integration limits per line by careful visual inspectionafter overplotting the entire time series. In this paper, we con-centrate on the equivalent width of the lines (the moment of or-der zero, hereafter abbreviated as EW) and on their heliocentriccentroid velocity (the first-order moment, denoted as (cid:104) v (cid:105) ). Typ-ical uncertainties for the centroid velocities due to uncertaintiesin the integration limits and the noise in the spectral line rangefrom 0.1 to 0.4 km s − , while the systematic uncertainty due tolimited knowledge of the laboratory wavelength can reach up toa few km s − . The latter is not of importance when studying timevariability of one and the same particular spectral line. The sys-tematic uncertainty is only of importance when comparing theaverage value of the centroid velocity for di ff erent spectral linesto interpret their amplitude as a function of the formation depthin the stellar photosphere or wind.The atomic data are not equally good for di ff erent spectrallines. In view of this, it is highly advantageous to make a line-by-line analysis per instrument and combine the line diagnosticvalues in Fourier domain rather than in the time domain. Indeed,merging in the time domain su ff ers from slight inconsistencies inDoppler velocities derived from adopted laboratory wavelengthsfor various lines and from imperfections in the normalisationof the continuum flux, while merging in the Fourier domain isnot a ff ected by that. Such an approach also allows to combinethe temporal variability from entirely di ff erent quantities derivedfrom photometry and spectroscopy, with the opportunity to re-veal low-amplitude variations that appear consistently in di ff er-ent types of independent data but with too low SNR in each ofthem separately to be significant.The method of combining data in the Fourier domain wasalready applied and illustrated by Aerts et al. (2006) to discoverpreviously unknown low-amplitude non-radial oscillation modes Article number, page 6 of 14. Aerts et al.:
Kepler photometry and high-resolution spectroscopy of the O9.5 supergiant HD 188209
Table 2.
Summary of the observations treated in this work; N is the total number of measurements and ∆ T is the total time base. Rayleigh is equalto 1 / ( ∆ T ) and Nyquist is set equal to 1 / (2 ∆ t ), where ∆ t is the minimal time between two consecutive exposures. Instrument HJD start HJD end N ∆ T Rayleigh Nyquist(-2450000) (-2450000) (d) (d − ) (d − ) Kepler , Mask 32 5276.4899 6424.0009 10085 1147.5110 0.00087 24.4738
Kepler , Mask 45 4964.5118 6424.0009 31074 1459.4891 0.00069 24.4738HERMES@Mercator 5726.4523 7526.7411 102 1800.2887 0.00056 26.5877FIES@NOT 5146.4053 5815.5758 11 669.1705 0.00149 8.2830T13 2m AST 6433.7684 6602.7555 228 168.9871 0.00592 46.9585SONG 7240.3758 7617.6022 417 377.2264 0.00265 47.8460
Fig. 6.
The centroid velocity (lower panel) and equivalent width (upper panel) of the entire spectroscopic data set for the He i = HERMES, green = FIES, blue = AST, and black = SONG. in the archetype monoperiodic β Cep star δ Ceti from the com-bination of MOST space photometry and high-resolution spec-troscopy. The rationale behind it is that the discrete Fourier trans-forms of di ff erent time series data of the same variable star allachieve a maximum at frequencies that are present in the data,while spurious frequencies due to noise or aliasing are di ff erentfor the various independent data sets. In this respect, it is partic-ularly useful to combine data sets with a di ff erent sampling rate,as is the case for the SONG and HERMES data. Multiplicationof the Fourier transforms of the di ff erent data sets, after normal- ising them according to the frequency of maximal amplitude,implies that true frequencies get higher multiplied dimensionlessamplitude while spurious frequencies tend to cancel each other.In the case of δ Ceti mentioned above, it concerned frequenciesof standing waves due to coherent heat-driven oscillation modes.However, the same principle also applies to isolated frequenciesof standing waves excited stochastically as in sun-like stars andred giants, or to frequencies belonging to an entire spectrum oftravelling waves. Whatever their excitation mechanism, standingwaves or travelling waves inside a star have frequency values
Article number, page 7 of 14 & A proofs: manuscript no. hd188209-astroph
Fig. 7.
Amplitude spectra for the EW of the He i ffi cient data (upper panel, the reddashed lines indicate four times the average noise level computed overthe range [0 ,
10] d − ). The lower panel shows the result of multiplyingthe periodograms in the upper panels after weighing them according tothe SNR of the dominant frequency and placing the dominant frequencyafter this multiplication at value 1. connected with the stellar structure properties and so they occurat the same values in the various data sets. He i We first considered the He i i ff erences occur in the EW of the line for the fourspectrographs but also for the same spectrograph during di ff er-ent epochs. This may in part be due to imperfect normalisation,but it may also be caused by the rather irregular variability of Fig. 8.
Same as Fig. 7 but for (cid:104) v (cid:105) of the He i the star, as revealed in the Kepler data in Fig. 3. Even though themoments in the definition by Aerts et al. (1992) are specificallydefined such as to compensate optimally for small di ff erences inthe EW (cf. lower panel of Fig. 6), it is advantageous to treat thefour (cid:104) v (cid:105) data sets separately, given that the four average valuesof this quantity per spectrograph is not equal and merging in thetime domain prior to the computation of the Fourier transformwould introduce spurious low frequencies.The individual Scargle periodograms (Scargle 1982) for theHERMES, SONG, and AST data sets of the EW and (cid:104) v (cid:105) of theHe i ff er from daily aliasing,a well-known phenomenon in time-series analysis of single-sitedata sets. The (cid:104) v (cid:105) based on the AST data do not lead to signif-icant frequencies (maximum amplitude typically between 3 and4 times the SNR), but the HERMES and SONG data sets do re-veal significant low-frequency amplitude, with a factor typicallybetween 10 to 12 times the SNR for SONG and 4 to 6 times the Article number, page 8 of 14. Aerts et al.:
Kepler photometry and high-resolution spectroscopy of the O9.5 supergiant HD 188209
Fig. 9.
Normalised periodogram for the tangential velocity of internalgravity waves occurring in 2D simulation D11 in Rogers et al. (2013),properly scaled in frequency to take into account the mass and evolu-tionary status of HD 188209 with respect to the simulated stellar model.
Fig. 10.
Multiplied Scargle periodograms after weighting according tothe SNR of the frequency with highest amplitude for the (cid:104) v (cid:105) (and EW inthe inset) of the He i Kepler photometry in Mask 32.
SNR for HERMES (Fig. 8). For the EW, the significance levelsare typically somewhat lower than for the (cid:104) v (cid:105) (Fig. 7).For each of these individual periodograms, the maximumamplitude was sought and transformed to its value expressed asa function of the average SNR, where the latter was computed inthe Scargle periodogram over the frequency range [10 ,
20] d − .The dashed lines in Figs 7 and 8 are placed at four times this SNRSubsequently, the three Scargle periodograms expressed in unitsof the SNR were multiplied, after which they were normalisedsuch that the maximum peak after this multiplication was placedat value 1.0. The results of this procedure are shown in the lowerpanels of Figs 7 and 8. It can be seen that this procedure leads toa far clearer view of the variability in this single spectral line, thealiasing and noise peaks in the individual three data sets beingsubstantially reduced in this way. Variability is found to occur at Fig. 11.
The centroid velocities based on di ff erent spectral linesavailable in the HERMES spectroscopy: black: He ii ii i i i iii β versus H γ . a multitude of low frequencies, in the range [0 ,
2] d − , entirelyin agreement with the Kepler scattered-light photometry.The morphology of the normalised periodogram in Fig. 8is unlike those encountered for B dwarfs pulsating multiperi-odically in coherent heat-driven gravity modes (e.g., De Cat &Aerts 2002). In view of this discrepancy and the similar ampli-tude spectrum obtained from the
Kepler photometry, we com-pare the frequency spectrum of HD 188209 in Fig. 8 with predic-tions from hydrodynamical simulations based on convectively-driven internal gravity waves. In order to do so, we follow a sim-ilar approach as in Aerts & Rogers (2015); only here we have asimpler case because the 2D simulations by Rogers et al. (2013)provide us with velocity information that allows a direct mean-ingful comparison with Fig. 8, while Aerts & Rogers (2015) hadto transform this information to brightness variations to be ableto compare with CoRoT light curves. Given that HD 188209 isevolved and about ten times more massive than the model of3 M (cid:12) adopted for the simulations by Rogers et al. (2013), weperformed a scaling with a factor 0.53 that occurs between thefrequencies of dipole gravity waves of a 3 M (cid:12)
ZAMS star andof a 30 ˙M (cid:12)
TAMS star (Shiode et al. 2013, Table 1). Moreover,we normalise the tangential velocity of the 2D simulations (theradial component being entirely negligible - see Fig. 1 in Aerts& Rogers (2015), right panel), to their highest amplitude, con-sidering simulation run D11 (a non-rigidly rotating star whosecore rotates 1.5 times faster than its envelope). The outcome ispresented in Fig. 9. While this exercise does not allow a peak-to-peak comparison of the frequencies in the two spectra, theoverall morphology in Figs 8 and 9 is similar.Next, we went one step further and considered a multi-plied Scargle periodogram deduced from the three spectroscopicHe i Kepler data in Mask 32 as a fourthindependent data set. Its highest amplitude peak has a signifi-cance of 137 times the SNR in the frequency range [10 ,
20] d − ,the average SNR being 27 ppm in that interval. The outcome ofthis multiplication is shown in Fig. 10. Figures 7, 8, and 10 allpoint towards the same conclusion: HD 188209 reveals signifi-cant variability with an entire spectrum of significant frequen-cies below 2 d − . The morphology in the periodograms point to-wards travelling waves at the origin of this variability, given the Article number, page 9 of 14 & A proofs: manuscript no. hd188209-astroph
Fig. 12.
The centroid velocity (left y -axis) and Kepler photometry in Mask 45 (right y -axis) during the only epoch where the space and ground-based data overlap in time. The colour coding is as follows: H β : black, H γ : light blue, He i i i i ii ii iv iii Kepler lightcurve is shown as dotted line while its mirrored version with respect to the average
Kepler magnitude is shown as the full line. density of frequency peaks and the absence of any clear relation-ship between the dominant peaks. Thus, we find fully compat-ible variability results between the
Kepler scattered-light spacephotometry and the He i We repeated the entire procedure described in the previous sec-tion for the other photospheric lines that are present in allfour spectroscopic data sets. It concerns He i i i ii ii iv iii Allthese lines behave fully consistently with the He i ff erence beingthe amplitudes of their EW and (cid:104) v (cid:105) time series (not the relativeamplitudes expressed in terms of the SNR, which are remark-able similar for all those lines). Figure 11 shows a comparison ofthe (cid:104) v (cid:105) -values for several lines available in the HERMES spec-troscopy, where we plot them line-by-line. It can be seen thatthere is large consistency among the values of the centroid veloc-ities. The values of (cid:104) v (cid:105) for the He i lines have the strongest sim-ilarity. The Si iii ff ected by the base of the wind than the He i β and He ii +
45 km s − with respect to all other spectral lines when con-sidering its laboratory wavelength of 4685.71Å. This anomalyoccurs for all four the spectroscopic data sets. It is known thatthe He ii ff ected by the temperature, wind density, metallicity, andamount of EUV flux. However, we cannot explain the detectedredward shift corresponding with some 0.7Å that we find forthe absorption profile of this line. We are aware of at least oneother star where this line is in absorption and for which the sameanomaly is reported in the literature (see Fig. 15 in Massey et al.(2004) for the O8.5 I(f) supergiant AV 469 in the Small Mag-ellanic Cloud, a star with quite similar fundamental parametersthan HD 188209 except for the metallicity). Recent laboratorymeasurements of He ii pointed out that its spectral line structurenear 4686Å is complex (Syed et al. 2012). Caution of its inter-pretation for absorption lines of hot evolved stars seems in order.The H β line of HD 188209 behaves very similarly to the H γ line, the latter only being suitable for our analysis in the HER- Article number, page 10 of 14. Aerts et al.:
Kepler photometry and high-resolution spectroscopy of the O9.5 supergiant HD 188209
Fig. 13.
The centroid velocity (left y -axis, bottom panel) and H α emission strength expressed in continuum units (left y -axis, top panel), as wellas the Kepler photometry in Mask 45 (right y -axes) during the only epoch where the space and ground-based data overlap in time. The The colourcoding is as follows: H α : light green, H β : black, H γ : light blue, He i Kepler light curve is shown asdotted line while its mirrored version with respect to the average
Kepler magnitude is shown as the full line.
MES data set. This is also the case for the H α and Si iii ii Kepler space photometry andground-based spectroscopy. It concerns data in Mask 45 andHERMES spectroscopy. The centroid velocities of all the avail-able spectral lines and the space photometry taken during theseeight days are shown in Fig. 12. The He ii α fall outside the boundaries of this plot. It is seen that all spectrallines show consistent behaviour in their centroid velocity in thetime domain .The He ii α , H β , and H γ ,and in comparison with the Kepler photometry, in Fig. 13. It canbe seen that He ii β , and H γ are fully in agreement with each other in terms of variability. The H α emission varies in an-tiphase with the H α centroid velocity. While the relation betweenthe spectroscopic variability and the Kepler photometry is hardto unravel from Figs 12 and 13, the morphology of the variousquantities in the time domain is similar.
The spectroscopic vari-ability of the wind and of the photosphere and the photometricvariability are all in agreement in terms of amplitudes and peri-odicities.
Among the spectroscopic data sets, the highest sampling rate isobtained with SONG. A 5 d excerpt of (cid:104) v (cid:105) derived from ninespectral lines available in these data is shown in Fig. 14. We findsimilar behaviour for most of the spectral lines. The figure re-veals di ff erent behaviour in variability from night to night, withdominant periodicities typically longer than 10 h and quite dif-ferent amplitudes per night. This confirms the earlier finding of Article number, page 11 of 14 & A proofs: manuscript no. hd188209-astroph variability with frequencies below 2 d − in the Fourier domain,but now by visual inspection in the time domain based on thespectral lines available in the SONG spectroscopy.Thanks to its construction, the HERMES spectrograph of-fers the best long-term stability of the four instruments usedhere. Indeed, its fiber-fed design connected with a temperature-controlled room and its dedicated calibration pipeline werespecifically defined and implemented with long-term monitor-ing of variable phenomena in mind (Raskin et al. 2011). Thisdata set of HD 188209 is therefore best suited to illustrate thelong-term spectroscopic variability of the star in the time do-main. We show the centroid velocities for the eleven availablespectral lines in the HERMES data for the first four epochs ofmonitoring in Fig. 15, where the HJD for the various spectrallines have been shifted along the x − axis of the plot with mul-tiples of 12 days for visibility purposes. The data for the firstepoch in Fig. 15 corresponds with that shown in Fig. 12. Fig-ures 12 and 15 illustrate the consistency between the (cid:104) v (cid:105) -valuesfor the various spectral lines, both on a time scale of a week andof two years.A summary of all the computed line diagnostic is providedin Table 3 for the HERMES data for which the centroid veloci-ties have been plotted in Fig. 15 (except for H α as these do notfit the plot). The colour coding adopted in Figs 12, 13, 14, and15 has been indicated and the lines are listed in order of increas-ing average centroid velocity, corresponding to increasing line-formation depth into the wind / photosphere. Despite unknownsystematic uncertainties for (cid:104) v (cid:105) due to limitations in the knowl-edge of the laboratory wavelengths (of order km s − ), we detecta radial gradient in the average centroid velocities for the di ff er-ent lines (Table 3), where, as already mentioned, the He ii (cid:104) v (cid:105) is very similar if one keeps in mind thatline blending a ff ects this range and is di ff erent for the variousspectral lines. Even though they are formed partly in the windin view of their more negative average velocity, the variabilityof H β and H γ behaves similarly to the one of the helium andmetal lines. Hence, the photospheric variability does not changeat the bottom of the stellar wind. Only the H α variability seemsdominated by the wind behaviour.
5. Discussion and conclusions
In this work, we have used the
Kepler spacecraft far beyond itsnominal performance by studying scattered-light photometry ofthe bright blue supergiant HD 188209 while it was situated inbetween active CCDs. Aperture photometry of its scattered lightdelivered the first four-years long uninterrupted high-cadencelight curve of a blue supergiant, reaching a precision of some27 ppm at frequencies above 10 d − . We found similarities in themorphology of the frequency spectrum derived from the scat-tered light and from line-profile diagnostics of several spectrallines in long-term ground-based high-resolution spectroscopy. Amajor conclusion of this work is that the range of detected fre-quencies in the space photometry and in the ground-based spec-troscopy is the same. All the frequency spectra point towardsvariability occurring in the photosphere and consistently prop-agating into the bottom of the stellar wind. The nature of theshort-time Fourier transforms of the high-cadence photometryexcludes an interpretation in terms of standing waves connectedwith non-radial gravity-mode oscillations, but rather points to-wards the excitation of an entire spectrum of travelling internalgravity waves triggered by core and / or envelope convection. This is a plausible explanation in terms of the measured velocity am-plitudes and the multitude of detected frequencies in the regimebelow 2 d − .While we could not find a simple point-to-point relation-ship between the photometric and spectroscopic data that weretaken simultaneously, we did find full consistency in the fre-quency range caused by these independent data sets. Moreover,the overall morphology of the variable patterns in the time do-main and in the frequency spectra in Fourier space derived fromthe scattered-light photometry and from the centroid velocitiesdeduced from the spectroscopy is very similar. The observedfrequency spectra of HD 188209 are in qualitative agreementwith those for the tangential velocities based on 2D hydrody-namical simulations of internal gravity waves in a massive star.We thus conclude to have found the first observational evidenceof the occurrence of such waves in a massive blue supergiant.Along with the discovery of such waves in young O-type dwarfs(Blomme et al. 2011; Aerts & Rogers 2015), we revealed at leastone important mechanism of angular momentum transport activein massive stars during and beyond core-hydrogen burning thatis currently not included in stellar evolution models. It remainsto be studied how much impact the omission of this ingredienthas for stellar evolution theory.The tangential velocities associated with internal gravitywaves in the stellar photosphere of massive stars are of order atenth of a km s − per individual wave (Rogers et al. 2013). It hasalready been shown by Aerts & Rogers (2015, their Fig. 5) thatthe collective e ff ect of hundreds of such internal gravity waveson line profiles is very similar to the e ff ect due to a collectionof coherent heat-driven gravity-mode oscillations, particularlyfor the line wings. Given that both coherent standing gravity-mode oscillations and running gravity waves have completelydominant tangential velocities (at the level that their radial ve-locity component is negligible), any proper modelling of macro-turbulent line broadening due to gravity waves as detected inHD 188209 requires fitting a tangential macroturbulent velocityfield to the spectral lines. Acknowledgements.
This project has received funding from the European Re-search Council (ERC) under the European Union’s Horizon 2020 researchand innovation programme (Advanced Grant agreements N ◦ ◦ / / under REA grant agreement N ◦ Kepler
Discovery mission was provided by NASA’sScience Mission Directorate. The authors gratefully acknowledge the entire
Ke-pler team, whose outstanding e ff orts have made these results possible. This re-search has made use of the SIMBAD database, operated at CDS, Strasbourg,France, and of the Multimission Archive at STScI (MAST), USA. We also ac-knowledge use of the Atomic Line List o ff ered by the University of Kentucky,USA and maintained by Peter van Hoof, Royal Observatory of Belgium. References
Aerts, C. 2015, Astronomische Nachrichten, 336, 477Aerts, C., De Cat, P., Handler, G., et al. 2004, MNRAS, 347, 463Aerts, C., de Pauw, M., & Waelkens, C. 1992, A&A, 266, 294Aerts, C., Lefever, K., Baglin, A., et al. 2010, A&A, 513, L11Aerts, C., Marchenko, S. V., Matthews, J. M., et al. 2006, ApJ, 642, 470Aerts, C. & Rogers, T. M. 2015, ApJ, 806, L33Aerts, C., Simón-Díaz, S., Catala, C., et al. 2013, A&A, 557, A114Bedding, T. R., Mosser, B., Huber, D., et al. 2011, Nature, 471, 608Bischo ff -Kim, A. & Østensen, R. H. 2011, ApJ, 742, L16 Article number, page 12 of 14. Aerts et al.:
Kepler photometry and high-resolution spectroscopy of the O9.5 supergiant HD 188209
Fig. 14.
Excerpt of the SONG spectroscopy during 5 consecutive days. The centroid velocity is shown for nine available spectral lines (theHe ii Table 3.
Summary of variability properties of the centroid velocity (cid:104) v (cid:105) and EW for twelve spectral lines of HD 188209 available in the HERMESdata, spanning 1800 d. Spectral Laboratory Colour Minimum Maximum Range Average Minimum Maximum RangeLine Wavelength (Å) Code (cid:104) v (cid:105) (km s − ) (cid:104) v (cid:105) (km s − ) (cid:104) v (cid:105) (km s − ) (cid:104) v (cid:105) (km s − ) EW (Å) EW (Å) EW (Å)H α + γ β iii + i + i + i + i + ii ii + iv + ii + + Notes. ( ) Retrieved from the Atomic Line List available at ∼ peter/newpage/. Systematic uncertainties in theaverage values of (cid:104) v (cid:105) are di ff erent for the di ff erent spectral lines and may amount to several km s − per spectral line. This is typically an order ofmagnitude larger than the statistical uncertainties for each of the (cid:104) v (cid:105) -values for a fixed spectral line. Blomme, R., Mahy, L., Catala, C., et al. 2011, A&A, 533, A4Briquet, M., Aerts, C., Lüftinger, T., et al. 2004, A&A, 413, 273Buysschaert, B., Aerts, C., Bloemen, S., et al. 2015, MNRAS, 453, 89Chaplin, W. J. & Miglio, A. 2013, ARA&A, 51, 353Charpinet, S., Van Grootel, V., Brassard, P., et al. 2014, in Astronomical So-ciety of the Pacific Conference Series, Vol. 481, 6th Meeting on Hot Subd-warf Stars and Related Objects, ed. V. van Grootel, E. Green, G. Fontaine, &S. Charpinet, 105De Cat, P. & Aerts, C. 2002, A&A, 393, 965Degroote, P., Aerts, C., Baglin, A., et al. 2010, Nature, 464, 259Degroote, P., Aerts, C., Michel, E., et al. 2012, A&A, 542, A88 Deheuvels, S., Ballot, J., Beck, P. G., et al. 2015, A&A, 580, A96Eaton, J. A. & Williamson, M. H. 2004, Astronomische Nachrichten, 325, 522Fossati, L., Schneider, F. R. N., Castro, N., et al. 2016, A&A, 592, A84Foster, H. M., Reed, M. D., Telting, J. H., Østensen, R. H., & Baran, A. S. 2015,ApJ, 805, 94Fullerton, A. W., Gies, D. R., & Bolton, C. T. 1996, ApJS, 103, 475Fullerton, A. W., Massa, D. L., & Prinja, R. K. 2006, ApJ, 637, 1025Grundahl, F., Fredslund Andersen, M., Christensen-Dalsgaard, J., et al. 2017,ApJ, in pressGrundahl, F., Kjeldsen, H., Christensen-Dalsgaard, J., Arentoft, T., & Frandsen,S. 2007, Communications in Asteroseismology, 150, 300
Article number, page 13 of 14 & A proofs: manuscript no. hd188209-astroph
Fig. 15.