Atmospheric escape from HD 189733b observed in HI Lyman-alpha: detailed analysis of HST/STIS September 2011 observations

0.36 MB·English

by  Vincent Bourrier

Astronomy & Astrophysics manuscript no. Atmospheric˙escape˙HD189733b (cid:13)c ESO 2013
January 28, 2013
i α
Atmospheric escape from HD 189733b observed in H Lyman- :
detailed analysis of HST/STIS September 2011 observations
V. Bourrier1,2 A. Lecavelier des Etangs1,2, H. Dupuy1,2, D. Ehrenreich3, A. Vidal-Madjar1,2,
3 G. H´ebrard1,2, G. E. Ballester4, J.-M. D´esert5, R. Ferlet1,2, D. K. Sing6, and P.J. Wheatley7
1
0
2 1 CNRS,UMR 7095, Institut d’astrophysiquedeParis, 98bis boulevard Arago, F-75014 Paris, France
n 2 UPMC Univ.Paris 6, UMR 7095, Institut d’Astrophysiquede Paris, 98bis boulevard Arago, F-75014 Paris, France
a 3 Observatoire astronomique de l’Universit´e deGen`eve, 51 Chemin des Maillettes, 1290 Sauverny,Switzerland
J 4 Lunarand Planetary Laboratory, University of Arizona, 1541 E. University Blvd., Tucson, AZ 85721-0063, USA
5 5 Harvard-Smithsonian Center for Astrophysics, 60 Garden Street, Cambridge, MA 02138
2 6 AstrophysicsGroup, School of Physics, University of Exeter, Stocker Road, ExeterEX4 4QL, UK
7 Department of Physics,University of Warwick, Coventry CV4 7AL, UK
]
P
E
ABSTRACT
.
h
p Observations of transits of the hot giant exoplanet HD189733b in the unresolved Hi Lyman-α line show signs of hydrogen
-
o escapingtheupperatmosphereoftheplanet.NewresolvedLyman-αobservationsobtainedwiththeSTISspectrographonboard
r the Hubble Space Telescope in April 2010 and September 2011 confirmed that the planet is evaporating, and furthermore
t discovered significant temporal variations in the physical conditions of its evaporating atmosphere. Here we present a detailed
s
a analysis of the September 2011 observations of HD189733b, when an atmospheric signature was detected. We present specific
[ methods to find and characterize this absorption signature of escaping hydrogen in the Lyman-α line, and to calculate its

false-positive probability, found to be 3.6%. Taking advantage of the spectral resolution and high sensitivity of the STIS
1
v spectrograph, we also present new results on temporal and spectro-temporal variability of this absorption feature. We also
0 report the observation of HD189733b in other lines (Siiii at 1206.5˚A, Nv at 1240˚A). Variations in these lines could be
3 explained either by early occultation by a bow-shock rich in highly ionized species, or by stellar variations.
0
6 Key words.planetary systems - Stars: individual: HD189733
.
1
0
3 1. Introduction and observations Vidal-Madjar et al. 2008). It was also confirmed by two
1
subsequent observations at low spectral resolution: a 5%
: 1.1. Evaporation of hot Jupiters
v absorption of the whole Lyman-α line was indeed mea-
Xi Although the phenomenon of evaporation had been an- sured using unresolved emission line flux during plane-
ticipated by Burrows & Lunine (1995) and Guillot et al. tary transits observed with the Space Telescope Imaging
r
a (1996), the existence of a large number of hot Jupiters Spectrograph (STIS) instrument (Vidal-Madjar et al.
surviving atmospheric escape suggested that evapora- 2004) and the Advanced Camera for Surveys (ACS) in-
tion processes should be modest. In this frame it strument (Ehrenreich et al. 2008) onboard the Hubble
came as a surprise to discover that the exoplanet Space Telescope (HST). An independent analysis of the
HD209458b was losing gas (Vidal-Madjar et al. 2003). low-resolutiondatasetusedbyVidal-Madjar et al.(2004)
Transit observations of the Lyman-α line showed ex- has confirmed that the transit depth in Lyman-α is sig-
cess absorption due to an extended cloud of neutral hy- nificantlygreaterthanthe transitdepthdue tothe plane-
drogen Hi and constrained the (Hi) escape rate with tary disk alone (Ben-Jaffel & Sona Hosseini 2010). These
a lower limit of 1010gs−1 (Vidal-Madjar et al. 2003; threeindependentobservationsshowasignificantamount
Vidal-Madjar & Lecavelier des Etangs2004).Thisdiscov- of gas at velocities exceeding the planet escape velocity,
ery has been challenged by Ben-Jaffel (2007), but the leading to the conclusion that HD209458b is evaporat-
apparent discrepancy has been resolved and the result ing. This conclusion is strengthened by the observations
obtained by Ben-Jaffel on this first data set strength- ofabsorptionintheOiandCiilines(Vidal-Madjar et al.
ens the evaporation scenario (Vidal-Madjar et al. 2004; 2004), which led to the identification of a blow-off escape
mechanism. This is also confirmed by recent observations
Send offprint requests to: V.B. (e-mail: [email protected])
2 V.Bourrier et al.: Atmosphericescape observed in Hi Lyman-α
of absorption in the Cii and Siiii lines using the HST sit observations of HD189733b made in April 2010 and
Cosmic Origins Spectrograph (Linsky et al. 2010). Note September 2011 with the UV channel of the repaired
that additional analyses of these observations are being HST/STIS (using the same methodology as employed by
made (Ballester & Ben-Jaffel, in prep.). A similar con- Vidal-Madjar et al. 2003 for HD209458b), and showed
clusion has been reached in the case of WASP-12b with that the escape from this planet is subject to significant
observationsofMgiiandnearUVbroad-bandtransitab- temporal variations. In this paper we present a detailed
sorption (Fossati et al. 2010). analysis of the STIS transit observations of HD189733b
Based on these observational constraints, several made in September 2011. We also show a new time-
theoretical models have been developed to explain and resolved spectral signature of the absorption observed at
characterize the evaporation processes (Lammer et al. this epoch by using time-tagged photon count data. We
2003; Lecavelier des Etangs et al. 2004, 2007, 2008b; also present the transit observed in other lines, such as
Baraffe et al. 2004, 2005, 2006; Yelle 2004, 2006; Siiii at 1206.5˚A and Nv at 1239˚A and 1243˚A.
Jaritz et al. 2005; Tian et al. 2005; Garc´ıa Mun˜oz
2007; Holmstr¨om et al. 2008; Stone & Proga 2009;
Murray-Clay et al. 2009; Adams 2011; Guo 2011). In 1.2. HD189733b
the models developed by Holmstr¨om et al. (2008) and
The very hot Jupiter HD189733b is located 19.3 parsecs
Ekenba¨ck et al. (2010), the observed hydrogen is pro-
away from Earth, with a semi-major axis of 0.03 AU and
duced through charge exchange between the stellar wind
an orbital period of 2.2 days. The planetary transits and
and the planetary escaping exosphere. Even in these last
eclipses canbe usedto probeits atmosphereandenviron-
models, neutral hydrogen escaping the planet is required
ment (e.g., Charbonneau et al. 2008; D´esert et al. 2009).
for charge exchange with hot protons of the stellar wind.
HD189733b orbits a bright main-sequence star, with a
Therefore, all models led to the conclusion that most of
magnitude V=7.7. This K2V star emits one of the high-
the EUV and X-ray energy input by the host star is used
est Lyman-α flux ever measured for transits observations
by the atmosphere to escape the planetary gravitational
(withthenoticeableexceptionof55Cnc;Ehrenreich et al.
potential(Owen & Jackson2012;fordetailedestimatesof
2012), which makes HD189733b a particularly good can-
the EUV and X-ray input, see Sanz-Forcada et al. 2011).
didate to study atmospheric evaporation.
With this idea in mind, Lecavelier Des Etangs (2007)
developed an energy diagram in which the potential Thanks to the proximity of HD189733, and the
energy of the planet is plotted versus the stellar energy large surface covered by the planet against its rela-
received by its upper atmosphere. For HD209458b, tively small star, extensive observational studies have
the escape rates derived from this diagram are not been made of this hot Jupiter. The planet has a
high enough for the evaporating exoplanet to lose a mass M =1.13 Jupiter masses (M ) and a radius
p Jup
significant amount of its mass (Ehrenreich & D´esert R =1.16 Jupiter radii (R ) in the visible (Bakos et al.
p Jup
2011). However, the nature of planets with smaller 2006; Winn et al. 2007). The planetary disk transit re-
masses and closer orbits could be significantly altered sults in a ≈2.4% occultation depth from visible to near
by evaporation, and these planets would end as plane- infrared (D´esert et al. 2009; Sing et al. 2011). The short
tary remnants (Lecavelier des Etangs et al. 2004). The period of the planet (2.21858 days) has been mea-
recently discovered CoRoT-7b (L´eger et al. 2009) and sured precisely (H´ebrard & Lecavelier Des Etangs 2006;
Kepler-10b (Batalha et al. 2011) could be examples Knutson et al. 2009). Spectropolarimetry has measured
of a new category of planetary remnants, which were the strength and topology of the stellar magnetic field,
proposed for classification as “chthonian planets” (see whichreachesupto40G(Moutou et al.2007;Fares et al.
Lecavelier des Etangs et al. 2004). 2010). Sodium has been detected in the planet atmo-
For a few years, the only observed evaporating planet sphere by both ground-based (Redfield et al. 2008) and
was HD209458b,and many questions remained to be an- space-based observations (Huitson et al. 2012). Using the
swered.WhatistheevaporationstateofotherhotJupiters HST/ACS, Pont et al. (2008) detected atmospheric haze,
and very hot Jupiters? Is evaporation a common process which is interpreted as Mie scattering by small parti-
amongsthotJupiters?Howdoestheplanetarysystemand cles (Lecavelier Des Etangs et al. 2008a). High signal-to-
stellarcharacteristicsinfluencetheescaperate?Somelight noiseHST/NICMOS/STISobservations(Sing et al.2009,
has been shed on these questions thanks to the discovery 2011) and HST/WFC3 (Gibson et al. 2012) have shown
of HD189733b (Bouchy et al. 2005), a planet transiting that the near-IR spectrum below 2 µm is due to haze
a bright and nearby K star (V=7.7). Among the stars scattering and/or water absorption (Swain et al. 2008;
harboring transiting planets, HD189733b belongs indeed see also Gibson et al. 2011), and Rayleigh scattering
to the second brightest at Lyman-α. Despite the failure might even extend to longer wavelengths (Pont et al.
of the HST/STIS instrument one year before the dis- 2012). The detection of an H O signature using tran-
2
coveryofHD189733b,Lecavelier Des Etangs et al.(2010) sit photometry (Tinetti et al. 2007) has been subjected
detected atmospheric escape through Lyman-α transit to controversy (Ehrenreich et al. 2007; Agol et al. 2010;
observations with the HST/ACS instrument. Moreover, D´esert et al.2009,2011).Ithasbeententativelyproposed
Lecavelier des Etangs et al. (2012) compared two tran- thatCO moleculescanexplainthe excessabsorptionseen
V.Bourrier et al.: Atmosphericescape observed in Hi Lyman-α 3
at 4.5 µm (Charbonneau et al. 2008; D´esert et al. 2009; 2. Observations
Knutson et al. 2012). Applying best-estimate approaches
The observationsofHD189733binthe Lyman-αlineper-
to dayside infrared emission spectra of HD189733b,
formed in 2007 - 2008 showed significant absorption in
Lee et al. (2012) and Line et al. (2012) both reportedthe
the exosphere of the planet, while in April 2010no atmo-
detection of H O and CO in the atmosphere (see also
2 2
sphericabsorptionsignaturewasdetected.Toaddressthe
Swain et al. 2009). Using Spitzer spectroscopy of plane-
question of these variations, new transit observations of
taryeclipses,Grillmair et al.(2008)alsofoundevidenceof
the planetweremadeinSeptember2011withthe G140M
H Oabsorptionsignaturesandpossiblyweather-likevari-
2
grating of the HST/STIS. The log of these time-tagged
ationsintheatmosphericconditions.NotethatAgol et al.
observations is given in Table 1. The analysis described
(2010) found an upper limit of 2.7% on the variability
in this paper was made with 1D spectra extracted us-
of the dayside planet flux, which rules out the most ex-
ing CALSTIS (version 3.32) data pipeline. These spectra
tremeweatherfluctuationsonHD189733b.Analternative
cover the far-UV wavelengths from 1195to 1248˚A with a
source for the observed variability may be the change of
spectral resolution of about 20kms−1 at 1215.67˚A. The
location of a dayside hot spot detected at an offset from
planet was observed during four consecutive orbits of the
thesubstellarpoint(Knutson et al.2007,Agol et al.2010,
HST around the Earth: two orbits before the transit, one
Majeau et al. 2012). A sensitive search with GMRT has
during the transit, and one after the transit. The first ex-
provided very low upper limits on the meter-wavelength
posurelasts1800sandthethreeothersabout2100s.Data
radio emission from the planet, indicating a weak plan-
acquisitions are interrupted by Earth occultation lasting
etary magnetic field (Lecavelier Des Etangs et al. 2009,
about3500s.ContrarytoACSobservations,forwhichthe
2011).
co-addition of three independent transits was necessary
to detect the transit signature in the Lyman-α line, the
highestsensitivityoftheSTISdataallowsanindependent
1.3. Previous Lyman-α observations of HD189733b study of each transit observation.
3. Data analysis
In 2007-2008, because of the failure of the Space
Telescope Imaging Spectrograph (HST/STIS) in 2004,
3.1. Resulting spectra
HD189733 was observed with the Solar Blind Camera of
the (HST/ACS). Transit observations showed a diminu- We identified in the STIS spectra the stellar emission
tion of 5.05±0.75% of the entire Hi Lyman-α curve lines of Siiii (1206.5˚A), Ov (1218.3˚A), the Nv doublet
(Lecavelier Des Etangs et al. 2010). This was more than (1242.8˚A and 1238.8˚A), and the bright Hi Lyman-α line
the 2.4% planetary disk occultation depth and was inter- (1215.67˚A) (Fig. 1). To calculate the radialvelocities rel-
preted as the result of the evaporationof the planet’s up- ative to the star, we need to estimate the radial velocity
per atmosphere. However, the limited spectral resolution of the star relative to the STIS wavelength calibration.
of the ACS data in the far-ultraviolet wavelengths of the Usingthe stellaremissionlineswefoundthatthestarhas
Lyman-α line (at 1215.67˚A) implies that no conclusion a red-shifted velocity of ∼5kms−1 (heliocentric).
couldbemadeontheradialvelocitiesoftheescapinggas, In order to detect possible transit signatures of
and this detection called for more observations. After the HD189733b and its atmosphere, we calculated for each
refurbishmentofHST inMay 2009,HD189733bwasthus stellar line the time evolution of the total emission flux,
observed during transit in April 2010 with the G140M or of the flux within a given wavelength range. We also
grating of the repaired HST/STIS. Disappointingly, no compared the April 2010 and September 2011 observa-
significant absorption was detected in the Lyman-α line tions of the stellar lines other than the Lyman-α line
(2.9±1.4%, see Lecavelier des Etangs et al. 2012), other (for comparison of this line between 2010 and 2011, see
than the occultation depth by the planetary disk alone. Lecavelier des Etangs et al. 2012). A signature of the at-
Note that Jensen et al. (2012) reported a significant ab- mosphereis consideredto be detectedwhen anexcess ab-
sorption feature at Hα in the transmission spectra of sorptionisseeninadditionto theabsorptionby theplan-
HD189733b. This transit-dependent absorption was de- etary disk itself during the planet transit.
tected within a narrow band at the line center (-24.1
to 26.6kms−1). Because of the limited resolution of the 3.2. Nv doublet (at 1238.8˚A) and Ov line
ACS spectra, Jensen et al. could not directly compare
the velocity ranges between their H-α absorption detec- For the April 2010 and September 2011 observations, we
tion and the Hi Lyman-α absorption feature reported by compared the total flux in the stellar Siiii emission line
Lecavelier Des Etangs et al. (2010). Although the stellar and the Nv doublet before, during, and after the transit.
Lyman-α emission line is strongly absorbed by the inter- ResultsaresummarizedinTable 2.Wedonotdetectany
stellarhydrogenovertheirobservedvelocityrange,ourre- variationshowingatransit-likesignalinthe2010orinthe
cent and highly resolved STIS observations could enable 2011 observations for the Nv line at 1238.8˚A. All varia-
a more direct comparison between the two atmospheric tions, in particular differences in flux between 2010 and
signatures. 2011,aremostlikelyattributabletostellarvariationsina
4 V.Bourrier et al.: Atmosphericescape observed in Hi Lyman-α
Data set Date Observation Time from center of transit
Start End Start End
HSTorbit #1 2011-09-07 19:49:21 20:20:55 -03:31:02 -02:59:28
HSTorbit #2 2011-09-07 21:18:37 21:56:10 -02:01:46 -01:24:13
HSTorbit #3 2011-09-07 22:54:26 23:31:59 -00:25:57 00:11:35
HSTorbit #4 2011-09-08 00:30:16 01:07:49 01:09:52 01:47:25
Table 1. Log of the September 2011 observations. Time is given in UT.
vations of an early ingress in the Mgii line (Fossati et al.
25 2010, Haswell et al. 2012). Alternatively, intrinsic stellar
line variations could be the explanation of our observa-
)
−1 Å 20 Lyman−α tions. This raisesthe questionof possible stellar line vari-
−2 m 15 acatisoenosftrtihgegeCreidi flbayrethdeeptelacnteedt,iansAalCreSadoybsseursvpaetciotneds oinf tthhee
c
−1 s sameplanet(Lecavelier Des Etangs et al.2010).TheSiiii
g 10 line and Nv doublet thus appear as possible candidates
r
e Si III O V N V to diagnose the star-planet interaction (SPI), and addi-
−14 0 5 tional observations are needed to address this possibility.
1 Hereafter we consider only the Lyman-α observations.
(
x
u 0
Fl
−5 April Si III September Si III
4 2010
1200 1210 1220 1230 1240 1250 2011
Wavelength (Å) −1 Å)
−2 m 3
c
Fig.1. Plot of the wavelength-calibrated STIS spectrum −1 s
ofHD189733bwiththeLyman-α,Siiii,Ov,andNvstel- erg 2
lar emission lines. −14 0
1
x (
u
linearisingfromanactiveregionofthestellaratmosphere. Fl 1
TheanalysisoftheOvline,fainterandlocatedwithinthe
Lyman-α line red wing, gave no significant results. 0
1206.0 1206.5 1207.0 1206.0 1206.5 1207.0
Wavelength (Å) Wavelength (Å)
3.3. Siiii line and Nv doublet (at 1242.8˚A) Fig.2. Plot of the flux in the Siiii line for the 2010
(leftpanel)and2011(rightpanel)observations.Theblack
Althoughthereisnosignificantvariationinthefirstepoch
observations for the Siiii line and the Nv doublet, in the dashed line shows the spectrum before the planet transit;
the green solid line with error bars at the 1σ level shows
secondepoch wesee significanttime variationsofthe flux
measuredin the Siiiiline, and in the Nv line at 1242.8˚A the flux during the transit; the red dotted line shows the
flux after the transit.
(seeTable 2).Inthe 2011observations,these stellarlines
are observed to be fainter before the transit than during
and after the transit (Fig 2 and Fig 3). There is a 3.5%
probability to find the high variations observed between
the fluxes before,during,andafter the transitfor the Nv 3.4. Lyman-α line
lineat1242.8˚A(seetheχ2 valuesinTable 2).Thisprob-
3.4.1. The stellar Lyman-α emission line profile
ability is 0.6% for the Siiii line. Variations in the Siiii
line brightness have already been observed in the plane- The Lyman-α line is the brightest stellar line in the STIS
tarysystemofHD209458b(Linsky et al.2010).However, spectra from 1195 to 1248˚A. The single stellar Lyman-α
here we see a higher flux during the transit, which can- line is massively absorbed by the interstellar atomic hy-
not be interpreted by a classical absorption in the plan- drogenHi in a narrowband atthe line center (1215.6˚A),
etary exosphere. Our observation could be interpreted and to a lesser extent by the interstellar deuterium Di
by an early ingress caused by the formation of a bow- around 1215.3˚A. As a result, the spectrally resolved line,
shock surrounding the planet magnetosphere. This was attheresolutionoftheG140Mspectrograph,iscomposed
modeled by Llama et al. (2011) and Vidotto et al. (2010, of two peaks separated by this deep absorption. In order
2011a,b), and explains for example the WASP-12b obser- to estimate the Lyman-α line flux and profile, we fitted
V.Bourrier et al.: Atmosphericescape observed in Hi Lyman-α 5
Siiiiline Nv doublet
Flux within 1206 - 1207˚A Flux within 1238.4 - 1239.1˚A Flux within 1242.5 - 1243.1˚A
(10−14ergs−1cm−2) (10−15ergs−1cm−2) (10−15ergs−1cm−2)
2010 2011 2010 2011 2010 2011
Before transit 1.05±0.04 0.87±0.04 3.59±0.19 2.77±0.16 1.74±0.13 0.97±0.10
Duringtransit 0.94±0.05 0.98±0.05 3.15±0.23 2.72±0.22 1.50±0.16 1.40±0.15
Aftertransit 1.07±0.06 1.08±0.06 3.07±0.23 2.77±0.22 1.75±0.17 1.19±0.14
χ2 3.82 10.1 3.86 0.04 1.65 6.70
Table 2.TotalfluxintheSiiiiemissionline(at1206.5˚A)andtheNvemissionline(at1242.8˚Aand1238.8˚A)before,
during, and after the transit. The χ2 values are the standard deviations associated to each stellar line in either 2010
or 2011.
8 April NV September NV Model χ2 k DOF BIC AIC
7 2010
2011 Reference model 36.1 9 40 71.1 54.1
−1 Å) 6 TISM6=0 35.8 10 39 74.7 55.8
−2 cm 5 D2oduiffbelereGntauVsosiigatnpLrSofiFles 3345..45 1121 3378 8718..13 5587..45
−1 s 1-Voigt profile 41.5 8 41 72.6 57.5
g 4
er 2-Gaussian profile 166.6 8 41 197.7 182.6
−15 10 3 Table 3. χ2, BIC and AIC for various models with k
ux ( 2 degrees of freedom used to fit the Lyman-α line profile.
Fl
1
0
1242.5 1243.0 1242.5 1243.0
Wavelength (Å) Wavelength (Å)
Fig.3.PlotofthefluxintheNvlinebetween1242.5and resulting profile shows a double peak emission, which is
1243.1˚A, for the 2010 (left panel) and 2011 (right panel) usuallyseeninthe Lyman-αline ofcoolstars,andcanbe
interpreted as self-absorption in the chromosphere of the
observations. The black dashed line shows the spectrum
star (Wood et al. 2005).
before the planet transit; the green solid line with error
barsatthe 1σ levelshowsthe flux during the transit;the In the search for the best model to fit the Lyman-α
red dotted line shows the flux after the transit. profile,weusedtheBayesianInformationCriterion(BIC)
and the Akaike Information Criterion (AIC) (see, e.g.,
de Wit et al. 2012). These criteria prevent us from over-
the line profile as observed in September 2011, following fitting and are based on the likelihood function given by
the method of Wood et al. (2005) and Ehrenreich et al. the χ2 and on a penalty term related to the number of
(2011). The best result was obtained by modeling the parameters in the fitting model (Crossfield et al. 2012;
stellar emission line using two Voigt profiles with the Cowan et al.2012).ThelowestBICisobtainedfortheref-
same width and damping constant. The free parameters erencemodeldescribedabove(seeparametersinTable3).
of the fit are the total stellar flux, the Voigt parameters When we allowed for the temperature of the ISM to be
(width, damping constant, and the wavelength difference a free parameter (allowing different turbulent widths for
betweenthetwoprofilescenters),andtheISMparameters the ISM hydrogen and deuterium absorption lines), we
(Hi column density, ISM and star radial velocities, and obtained a better χ2 but the larger BIC and AIC show
turbulence velocity). We used a D/H ratio of 1.5×10−5 that this decrease cannot be interpreted in terms of the
(e.g., H´ebrard& Moos 2003, Linsky et al. 2006). Finally information content of the data. We also obtained a bet-
weaddedasinglefreeparameterto fitthe strengthofthe terfittothedatabyusingatwo-GaussianLSFwiththree
wing of the line spreadfunction (LSF), which is added to additional free parameters, but again the larger BIC and
the G140M published tabulated LSF1. The wing of this AIC shows that the increase of freedom is responsible for
LSF is taken as a Gaussian with a fixed width measured the apparent improvement in the fit. We also concluded
inthepublishedG140MLSFtobeσ =4.6pixels. thatthebetterfittotheprofileobtainedbyusingtwodif-
WingLSF
The comparison between the best model reconstructed ferent Voigt profiles (with different widths and damping
Lyman-α stellar profile and the observedspectrum yields constants)isnotsignificant.Wealsotestedifmoresimple
a χ2 of 36.1 for 49 data points in the wavelength range emission line profiles with lower degrees of freedom could
1214.25-1215.5˚A and 1215.8-1217.1˚A (see Fig. 4). The be used (one single-Voigt profile or two-Gaussian profile)
and concluded that the reference model with two similar
1 http://www.stsci.edu/hst/stis/performance/spectral_reVsooilguttipornofiles provides the best fit to the data.
6 V.Bourrier et al.: Atmosphericescape observed in Hi Lyman-α
0.25
0.20
unt 0.15
o
c
n
a
e 0.10
M
0.05
0.00
−600 −400 −200 0 200 400 600
Velocity (km s−1 )
Fig.5. Plot of HD189733b Lyman-α line profile (black
line)superimposedwiththegeo-coronalemission(redand
Fig.4. Plot of the theoretical profile of HD189733 blue lines). The red line correspondsto the 2010 observa-
Lyman-α line. The black thin line shows the theoretical tions, the blue line to the 2011 observations. All lines are
intrinsic stellar emission line profile as seen by hydrogen averagedover the four HST orbits.The geo-coronalemis-
atomsescapingtheplanetaryatmosphere.Theblackthick sion is noticeably low during the 2011 observations.
line showsthe resulting profile after absorptionby the in-
terstellar hydrogen (1215.6˚A) and deuterium (1215.3˚A).
The line profile convolved with the HST G140M instru-
by the telescope ”breathing”, which is known to affect
mental line spread function (red line) is compared to the
STIS observations in the UV and in the visible because
observations (blue histogram), yielding a good fit with a
of variations of the telescope throughput. The breathing
χ2 of 36.1 for 40 degrees of freedom.
is caused by changes in temperature that the HST ex-
periences during each orbit. This causes small motions
of the secondary mirror in the STIS bench, which alters
3.4.2. Geo-coronal emission
the alignmentof the targetin the aperture (Kimble et al.
In the raw data, the stellar emission line is superimposed 1998). It may also change the focus position and the size
with the geo-coronal airglow emission from the upper at- of the point spread function, affecting the spectrograph
mosphere of the Earth (Vidal-Madjar et al. 2003). This throughput through the narrow 0.1” slit. Although the
emissionline canbe wellestimatedandremovedfromthe breathing depends on a large number of parameters and
finalspectrumusingthecalstisdatapipeline(version2.32 cannot be predicted, it can be modeled for a past ob-
ofNovember5,2010).Independentre-analysisofrawdata servation. Using the STScI Focus model (Cox & Niemi
using the same methodology as in Vidal-Madjar et al. 2011), we thus assessed the repeatability of the breath-
(2003) and D´esert et al. (2004) confirmed that the air- ing effect on the focus position during the 2011 obser-
glow emission can be efficiently subtracted and included vations. While the time windows of these observations
in the final error budget. Moreover, because we used a correspond to significant variations of the focus position,
narrow spectrograph slit of 0.1”, the airglow contamina- the 2010 observations coincide with times when the fo-
tionislimitedtothecentralpartoftheLyman-αlineand cus position is subjected to only low variations (Fig. 6).
does not contaminate the part of the spectrum where at- The shape and amplitude of these systematic variations
mosphericsignaturesaredetected,inthewingsoftheline can change between visits of the same target, as was
(Sect. 4). The airglow emission may be affected by daily found in STIS optical transit observations of HD209458
and seasonal variations, and the data of September 2011 (Brown et al. 2001; Charbonneau et al. 2002; Sing et al.
present a noticeably low airglow emission level (Fig. 5). 2008) and of HD189733 (Sing et al. 2011; Huitson et al.
2012), and in STIS FUV transit observations of 55Cnc
(Ehrenreich et al. 2012). Extensive experience with opti-
3.4.3. Telescope ”breathing”
cal STIS data over the last decade furthermore indicates
Using time-tagged acquisitions of the STIS spectra, we that the orbit-to-orbit variations within a single visit are
checkedvariationsatshorttimescaleswithinagivenHST both stable and highly repeatable, except for the first or-
orbitin asimilar mannerasdone by Ben-Jaffel(2007). In bit,whichdisplaysadifferenttrend(seeBrown et al.2001
the 2011 observations, we see a periodic variation of the and Charbonneau et al. 2002). Because of its repeatabil-
measured Lyman-α flux in phase with the HST position ity, this effect is easy to correct for.
on its orbit, with the same amplitude of about 10% at all For eachHST orbit,we divided the time-taggedexpo-
wavelengths and no spectral signature. These variations sures into six 379 s subexposures. Following the method
arenotseenatasignificantlevelinthe2010observations. of Sing et al. (2011) and Brown et al. (2001), we phase-
TheyareinterpretedbytheSTScISTISexpertsascaused folded these exposures over the HST-orbital period of
V.Bourrier et al.: Atmosphericescape observed in Hi Lyman-α 7
fit at the same HST orbital phase. This correction does
6.00
ns) 024...000000 nTthoheteiraspprpeeslcyptertacotaitvrheeeerfiexcrpsootnssuotrrrbeusictt(beFdeicgfa.our8s)ee.aoNcfhoitotesrbtdihitffabetyrwenceotaotdbrdetnainidng.
micro −2.00 the same estimates for absorptions and ratios measured
n( −4.00 using the full HST orbit exposures, with or without the
o
ositi 6.00 correction,asdonebyLecavelier des Etangs et al.(2012).
p
us 4.00 However, the correction must be taken into account for
Foc 2.00 measurements on partial orbit exposures obtained using
−−042...000000−4 −3 −2 −1 0 1 2 3 time-tagged data.
Time (hours) 1.15
Fig.6. Longitudinal motion of the secondary mirror for 1.10
the 2010 observations (top panel) and the 2011 observa-
tions (bottom panel). Time is given relative to the center d flux 1.05
oftheplanetarytransit.Hatchedzonesshowthetimewin- e
z 1.00
dows of our observations of HD189733b. ali
m
or 0.95
N
96 minutes. Each exposure was then spectrally summed 0.90
overtheentireLyman-αline,andnormalizedbythemean
0.85
valueofalltheexposuresofitsrelatedorbit.Weexcluded
−4 −3 −2 −1 0 1 2 3
the first orbit. We also excluded observations obtained Time (hours)
during the planetary transit. We found that the breath-
ing effect was best fitted with a second-order polynomial Fig.8. Raw fluxes not corrected for the breathing effect
(Fig. 7; see Ehrenreich et al. 2012 for a similar correction (empty squares), as a function of time in the September
of this effect). 2011 observations. Time is given relative to the center of
the planetary transit. Circles correspond to the corrected
flux. Each point stands for a 379 s exposure spectrum,
summed over the whole Lyman-α line, and normalized
by the flux over the entire corresponding orbit. Vertical
1.1 dashed lines show the beginning and end of ingress and
egress of the transit.
x
u
d fl 1.0
e
z
ali
m
or 4. Atmospheric hydrogen
N 0.9
4.1. Full Lyman-α flux
0.8 In a first step, to search for the transit signature of the
−0.1 0.0 0.1 0.2 0.3 0.4 planetary atmosphere in the Lyman-α line, we compared
HST orbital phase
the total line flux measured during the transit with the
fluxmeasuredbeforethetransit.Thetransitdepthbythe
Fig.7. Lyman-α fluxes for 379 s exposures, phase-folded
planetary disk alone as seen from the visible to the near-
on the HST orbital period. The empty circles correspond
infrared amounts to 2.4% (Pont et al. 2007; Sing et al.
to the first orbit. The green circles correspond to the
2009;D´esert et al.2009).TheobservationsofHD189733b
measurementsobtainedduring the planetary transit.The
made in April 2010 showed no significant excess in the
black line is the second-order polynomial fit to the other
transitabsorptiondepth, with a depth for the flux within
orbits measurements (filled black dots).
the whole Lyman-α line of 2.9±1.4%, including the plan-
etarydiskoccultation(Lecavelier des Etangs et al.2012).
Fig.7showsthatthefirstorbitseemstobelowerthan In contrast, a transit absorption of 5.0±1.3% is found
the expected unabsorbed Lyman-α stellar light curve. in the total flux of the Lyman-α line for the 2011 ob-
This could mean the absorption signatures described in servations. We fitted the flux measured during the tran-
Sect. 4, and obtainedwith the two orbits before the tran- sit with a classical planetary occultation model from
sit as reference, may be underestimated. We correct the Mandel & Agol(2002). The bestfit is consistentwith the
data for the breathing effect by dividing the spectra of presence of an extended absorbing atmosphere surround-
each 379 s subexposure by the value of the polynomial ingtheplanet(seeFig.9).Substractingtheplanetarydisk
8 V.Bourrier et al.: Atmosphericescape observed in Hi Lyman-α
absorption,the signature measuredduring the 2011 tran-
sitcorrespondstoanexcessabsorption,duetoatmosperic
25
hydrogen only, of 2.3±1.4%. This is consistent with the −1 Å)
excess absorption of 2.71±0.75% due to hydrogen only, −2 m 20
obtained with the HST/ACS observations in 2007-2008 c
(Lecavelier Des Etangs et al. 2010). −1 s 15
g
er
−14 0 10
1
x (
1.04 Flu 5
1.02 0
ux −600 −400 −200 0 200 400 600
ed fl 1.00 Velocity (km s−1 )
aliz 0.98
m
Nor 0.96 Fig.10. Lyman-α line profile. Four spectra, correspond-
ingtothefourHSTorbitsoftheSeptember2011observa-
0.94
tions,aredisplayedasafunctionofradialvelocityrelative
0.04
to the star. The black and blue lines show the fluxes be-
als 0.02
du 0.00 fore the planet transit; the green line with error bars at
Resi −0.02 the1σ levelshowsthefluxduringthetransit;theredline
−0.04 shows the flux after the transit.
−4 −3 −2 −1 0 1 2
Time (hours)
9 24
Fig.9. Plot of the total flux within the whole Lyman-α
8 23
line, as a function of time (blue square symbols). Each
HST orbit is divided into two 1127 s independent subex- −1 Å) 7 −1 Å) 22
posures, corrected for the breathing effect (except for the −2 m 6 −2 m 21
firstorbit,see Section3.4.3). Time is givenrelativetothe −1 s c −1 s c
centerofthe planetarytransit.Verticaldashedlines show g 5 g 20
er er
the beginning and end of ingress and egress of the tran- −14 0 4 −14 0 19
sit. Horizontal error bars centered on the symbols show 1 1
x ( x (
the duration of the exposures. A light curve fitted with u 3 u 18
Fl Fl
a classical planetary occultation model is displayed as a
2 17
solidblackline,withatransitdepthinexcesscomparedto
1 16
thetransitdepthobservedatopticalwavelengths(dashed
−300 −250 −200 −150 −100 50 100
black line). The bottom panel shows the corresponding Velocity (km s−1 ) Velocity (km s−1 )
residuals (empty symbols).
Fig.11.Zoomonthe blue wing (left panel)andredwing
(right panel) of the Lyman-α line profile displayed in
Fig. 10. The black and blue dashed lines show the fluxes
before the planet transit; the green solid line with error
4.2. Spectrally resolved absorption features barsatthe 1σ levelshowsthe flux during the transit;the
red dotted line shows the flux after the transit.
In contrast to the ACS spectra, the spectrally resolved
STIS spectra allow us to look for absorption signatures
within specific wavelength intervals (Fig. 10). Two clear
4.3. Quantitative estimates
absorptionregionscanbeseenwithintheLyman-αlinein
the2011transitobservation.Themostsignificantabsorp- To quantitatively characterize the absorption features
tion signature is visible in the blue part of the line from within the Lyman-α line, we studied the ratio of the flux
-230kms−1 to-140kms−1.Anotherfeature,althoughless during transit to the flux before transit. The transit sig-
significant, is visible at the peak of the red wing from nature is the combination of the absorption by the at-
60kms−1 to110kms−1 (Fig.11).Thereseemstobesome mospheric neutral hydrogen and the 2.4% occultation of
absorption within the same velocity region in the blue the stellar disk by the planetary disk at all wavelengths.
wing for the post-transit observation, albeit with a lower To characterize the absorption due to the hydrogen only,
depth. The red region displays no absorption feature in we decreased the flux before the transit by a factor cor-
the post-transit observation. In the following sections we responding to the planet absorption. Then, we searched
describe the methods used to quantitatively characterize for the strongest signature at every possible wavelength
these two signatures, in particular their precise velocity range,excludingintervalsnarrowerthan0.1˚A(abouttwo
range and absorption depths. pixels in the STIS spectra). We looked for the most sig-
V.Bourrier et al.: Atmosphericescape observed in Hi Lyman-α 9
nificant absorption, characterized by the highest value of 200kms−1. We find the most significant transit absorp-
the ratio of the relative absorption to its noise (Fig. 12). tion signaturein the range60 to 110kms−1 (see Fig. 13).
In the 2011 transit observations, the most significant Substracting the occultation by the planetary disk, this
signature is found within the range -230 to -140kms−1, signature yields a 5.5±2.7% absorption depth (2.0σ de-
yielding an Hi absorption depth of 12.3±3.7% (3.3σ de- tection). This detection is the same if we exclude the first
tection).Thissignature,duetothe atmospherichydrogen orbit, whether the spectra are correctedfor the breathing
only, is the most significant over the entire Lyman-α pro- effect or not. Cumulative transit depths due to the plan-
file. The measured velocity of the gas is higher than the etary disk and the atmospheric hydrogencan be found in
escape velocity of 60kms−1 from the planet (the effec- Table 4.
tiveescapevelocityisevenlowerthan60kms−1 athigher
altitude where we observe the absorption). The total ab-
sorptiondepthof14.4±3.6%,whichincludestheplanetary
disk occultation, also corresponds to high altitudes: it is
25
equivalent to a disk with a radius of 2.8 Jupiter radii.
Therefore,theobservedneutralhydrogenmustbeathigh −1Å) 20
−2m
acllutidtuedtehse,fierssctapHinSgTtohrebiptlainnetthegrcaavlictuyl.aNtiootneotfhtahteifflwuxebexe-- Flux −1gsc 15
er 10
fore the transit, the absorption depth is the same within −140
1σ (18.3±5.6%),butthedetectedrangeissomewhatnar- (1 5
rower,from -230 to -180kms−1. o 2.0
on tatio
ptie r 1.5
sorois
bn 1.0
A
8.0
−1Å) 1.00
−2m 6.0 n
Flux −1ergsc 4.0 sorptio 0.95
−140 2.0 Ab 0.90
1
(
on toatio 3.0 50 10V0elocity (km s−1 )150 200
ptie r
sorois 2.0 Fig.13.Same as inFig.12in the redpartofthe Lyman-
bn
A α line. The red dotted line is for ratios calculated over
domainsbetween60kms−1 andincreasingvelocities.The
n 0.9 greendash-dottedlineisforratiosbetween110kms−1and
o
pti 0.8 decreasing velocities.
or
s
b 0.7
A
0.6
−350 −300 −250 −200 −150 −100 −50 We also divided each exposure of a given HST orbit,
Velocity (km s−1 )
corrected for the breathing effect, into two independent
spectra with 1127 s exposure time. We measured the ab-
Fig.12. Plot of the flux (upper panel), absorption (bot-
sorption over the range -230 to -140kms−1 for the April
tompanel),andratiooftheabsorptiontoitsnoise(middle
2010 and September 2011 observations. A light curve fit-
panel) inthe blue partof the Lyman-αline, as a function
ted to the data using a model of a single planetary disk
of velocity. The upper panel displays the flux before the
occultation unambiguously shows that, in contrast to the
transit multiplied by the occultation factor of the plane-
2010 measurements, the 2011 flux measurements require
tary disk (black dashed line), to be compared to the flux
the presence of an extended hydrogen atmosphere. Each
during the transit (blue solid line). In the lower and mid-
of the two independent transit measurements of 2011 are
dle panels, total cumulative values are calculated within
consistentwitha14.4±3.6%absorption,interpretedasat-
domainsofincreasingrange.The reddottedline is forra-
tios calculated within domains between -230kms−1 and mosphericabsorptioninadditiontotheplanetarydiskoc-
cultation(Fig.14).Notethattheabsorbingatmosphereis
increasing velocities. The green dash-dotted line is for ra-
tios between -140kms−1 and decreasing velocities. likely not spherical, which explains why the post-transit
data of the 2011 observations cannot be fitted with the
simple model used here (see Section 4.6). A full descrip-
The same method was applied to the red wing of the tion of the geometry of the hydrogen cloud is beyond the
Lyman-α line, with a search window in the range 40 to scope of the present paper, however.
10 V.Bourrier et al.: Atmosphericescape observed in Hi Lyman-α
Transit depths HST/ACS (2007 - 2008) HST/STIS (2010) HST/STIS (2011)
not spectrally resolved spectrally resolved spectrally resolved
Planetary disk occultation 2.4% 2.4% 2.4%
Whole Lyman-αline 5.1±0.8% 2.9±1.4% 5.0±1.3%
-230 to -140kms−1 0.5±3.8% 14.4±3.6%
60 to 110kms−1 5.8±2.6% 7.7±2.7%
Table 4. Transit depths in a given wavelength range (including planetary disk occultation and atmospheric absorp-
tion), for the three observations of HD189733b.
1.10 April 2010 1.10 September 2011
ux ux
d fl 1.00 d fl 1.00
e e
aliz aliz
m m
Nor 0.90 Nor 0.90
0.80 0.80
als 0.1 als 0.1
du 0.0 du 0.0
esi esi
R −0.1 R −0.1
−4 −3 −2 −1 0 1 2 −4 −3 −2 −1 0 1 2
Time (hours) Time (hours)
Fig.14.Plotofthefluxbetween-230and-140kms−1inthebluewingoftheLyman-αlineasafunctionoftimerelative
to the center of the planetary transit(filled symbols).The red triangularsymbols are for the 2010observations,while
thebluesquaresymbolscorrespondtotheobservationsof2011.EachHSTorbitisdividedintotwo1127sindependent
subexposures, corrected for the breathing effect (except for the first orbit, see Section 3.4.3). Vertical dashed lines
show the beginning and end of ingress and egress of the transit. Horizontal error bars centered on the symbols show
the duration of the exposures. The light curve of the planet transit at optical wavelengths is displayed as a dashed
black line. A light curve fitted with a classical planetary occultation model to the flux during the transit is displayed
as a solid black line, and shows that the excess absorption feature detected in the 2011 data is not seen in the 2010
data. The bottom panel shows the corresponding residuals (empty symbols).
4.4. False-positive probabilities cludedfalse-positivesignaturesfoundonwavelengthinter-
vals covering1 or 2 pixels, which would be narrowerthan
Toassessthe significanceoftheabsorptionfeaturesfound the STIS resolution.The false-positive probability to find
in the Lyman-α light curve, we made the hypothesis that anexcessabsorptiondepthatmorethan3.3σintherange
the transit spectrum shows no absorption and calculated -350to-50kms−1isonly3.6%.Itisunlikelythatthispar-
withabootstrap-likemethodthefalse-positiveprobability ticular signature comes from statistical noise in the data.
to have signatures as significant as the one we detected, Thefalse-positiveprobabilitytofindanexcessabsorption
but caused by noise only. depth at more than 2.0σ in the range 40 to 200kms−1
is 24.6%. With this higher probability, the 5.5±2.7% ab-
Fromareference-averagedspectrum,wegeneratedtwo
sorption depth detected in the red wing is possibly due
random spectra simulating spectra before and during the
to noise in the data. Note, however, that a similar signa-
transit free of atmospheric absorption. The two spectra
ture was also observed in the case of HD209458b, with
were obtained by adding random Gaussian noise to the
an absorption depth of 5.2±1.0% in the red wing be-
reference spectrum, with an amplitude corresponding to
tween 1215.89 and 1216.43˚A (Vidal-Madjar et al. 2003,
the error estimated for the spectra before and during the
Vidal-Madjar et al. 2008).
transit.Wegeneratedthesetwospectrafor1,000,000runs,
andineachrunusedtheabsorptiondetectionmethodde-
tailedinSection4.3tofindthemostsignificantabsorption
4.5. Spectral and temporal variability
signature. Due to a better consideration of the searched
velocity intervals, we found lower estimates of the false- 4.5.1. Absorption depth as a function of radial velocity
positive probability than in Lecavelier des Etangs et al.
(2012). Indeed, taking into account that the range of Here we address the possibility to resolve the absorp-
the searched-for absorption feature cannot be narrower tion feature in the blue range of the Lyman-α line profile
than the spectral resolution of the spectrograph, we ex- between -230 and -140kms−1. Although the absorption