HST/NICMOS detection of HR 8799 b in 1998

Abstract

Three planets have been directly imaged around the young star HR 8799. The planets are 5-13 Mjup and orbit the star at projected separations of 24-68 AU. While the initial detection occurred in 2007, two of the planets were recovered in a re-analysis of data obtained in 2004. Here we present a detection of the furthest planet of that system, HR 8799 b, in archival HST/NICMOS data from 1998. The detection was made using the locally-optimized combination of images algorithm to construct, from a large set of HST/NICMOS images of different stars taken from the archive, an optimized reference point-spread function image used to subtract the light of the primary star from the images of HR 8799. This new approach improves the sensitivity to planets at small separations by a factor of ~10 compared to traditional roll deconvolution. The new detection provides an astrometry point 10 years before the most recent observations, and is consistent with a Keplerian circular orbit with a~70 AU and low orbital inclination. The new photometry point, in the F160W filter, is in good agreement with an atmosphere model with intermediate clouds and vertical stratification, and thus suggests the presence of significant water absorption in the planet's atmosphere. The success of the new approach used here highlights a path for the search and characterization of exoplanets with future space telescopes, such as the James Webb Space Telescope or a Terrestrial Planet Finder. Comment: ApJL, in press

Full text

Page 1

arXiv:0902.3247v1 [astro-ph.EP] 18 Feb 2009
DRAFT VERSION FEBRUARY 18, 2009
Preprint typeset using LATEX style emulateapj v. 08/13/06
HST/NICMOS DETECTION OF HR 8799 b IN 1998
DAVID LAFRENIÈRE1, CHRISTIAN MAROIS2, RENÉ DOYON3, TRAVIS BARMAN4
Draft version February 18, 2009
ABSTRACT
Three planets have been directly imaged around the young star HR 8799. The planets are 5–13 MJupand
orbit the star at projected separations of 24–68 AU. While the initial detection occurred in 2007, two of the
planets were recovered in a re-analysis of data obtained in 2004. Here we present a detection of the furthest
planet of that system, HR 8799 b, in archival HST/NICMOS data from 1998. The detection was made using
the locally-optimized combination of images algorithm to construct, from a large set of HST/NICMOS images
of different stars taken from the archive, an optimized reference point-spread function image used to subtract
the light of the primarystar fromthe images of HR 8799. This new approachimprovesthe sensitivity to planets
at small separations by a factor of ∼10 compared to traditional roll deconvolution. The new detection provides
an astrometry point 10 years before the most recent observations, and is consistent with a Keplerian circular
orbit with a ∼70 AU and low orbital inclination. The new photometry point, in the F160W filter, is in good
agreement with an atmosphere model with intermediate clouds and vertical stratification, and thus suggests the
presence of significant water absorption in the planet’s atmosphere. The success of the new approachused here
highlights a path for the search and characterization of exoplanets with future space telescopes, such as the
James Webb Space Telescope or a Terrestrial Planet Finder.
Subject headings: planetary systems — techniques: image processing
1. INTRODUCTION
After more than a decade marked by the great success of
the radial velocity and transit planet detection techniques,
the direct imaging technique has finally made its grand en-
try on the scene in late 2008 through a series of exciting
exoplanets discoveries (Marois et al. 2008; Kalas et al. 2008;
Lafrenière et al. 2008; Lagrange et al. 2008), including the
spectacular discoveryof the multiple-planetsystem HR 8799.
This system showcases three planets of mass 5–13 MJupor-
biting at projected separations of ∼24, 38 and 68 AU. It is
located 39.4 pc away from the Sun and appears to be seen
nearly face-on. The planets all orbit the star in the same di-
rection and likely in the same plane, consistent with forma-
tion within a circumstellar disk. The host is a 30–160 Myr-
old star of spectral type A5, mass ∼1.5 M⊙and luminosity
∼4.9 L⊙; it is also classified as both γ Dor and λ Boo types.
A large infrared excess is detected at ∼100 µm, suggesting
that the planetary system is surrounded by a large dust disk.
The reader is referred to Marois et al. (2008) and references
therein for further details on this system.
As the first multiple-planet system directly imaged,
HR 8799 offers many new possibilities to advance our un-
derstanding of planets. First, multi-wavelength photometry
and spectroscopy of these three planets, which are almost cer-
tainly of the same age and metallicity, will be of great value
for the validation and calibration of both evolutionary and at-
mospheric models of giant planets. Also, with good measure-
ments of the orbits of the planets, it could be possible to inde-
Electronic address: lafreniere@astro.utoronto.ca
1Department of Astronomy and Astrophysics, University of Toronto, 50
St. George Street, Toronto, ON, M5S 3H4, Canada
2National Research Council Canada, Herzberg Institute of Astrophysics,
5071 West Saanich Road, Victoria, BC, V9E 2E7, Canada
3Département de physique and Observatoire du Mont Mégantic, Univer-
sité de Montréal, C.P. 6128 Succ. Centre-Ville, Montréal, QC, H3C 3J7,
Canada
4Lowell Observatory, 1400 West Mars Hill Road, Flagstaff, AZ 86011,
USA
pendently constrain their masses through dynamical studies,
as already attempted by Fabrycky & Murray-Clay (2008).
Besides its high scientific interest, the discovery of the
HR 8799 system represents a great technical achievement
given the brightness ratios and angular separations between
the planets and the star. The main obstacle to direct imaging
of exoplanets, for both space- and ground-based telescopes,
is the scattering of stellar light by irregularities on optical
surfaces which creates a halo of bright quasi-static speckles
around the stellar core, masking the underlying fainter plan-
ets. The planets aroundHR 8799were initially discoveredus-
ingadaptiveoptics and the angulardifferentialimaging(ADI)
technique (Marois et al. 2006), which uses the natural field-
of-view rotation of an alt-az telescope to discriminate plan-
ets from quasi-static speckles. Effectively, this techniques al-
lows to construct a high-fidelity image of the stellar point-
spread function (PSF) that does not contain the signal of any
eventual planets; this reference PSF image is subtracted from
the target image to remove the light from the star while pre-
serving that of any eventual planet. This technique, coupled
with the locally-optimized combination of images (LOCI) al-
gorithm (Lafrenière et al. 2007b), has proved to be the most
efficient approach to detect planets from the ground (see e.g.
Lafrenière et al. 2007a).
Following the initial detection of HR 8799 b and c in Oc-
tober 2007, a careful re-analysis of adaptive optics observa-
tions of HR 8799 obtained in July 2004, using improved PSF
subtraction algorithms, allowed to recover both planets, thus
providing a baseline of four years relative to the most re-
cent observations of September 2008. This four-year baseline
firmly establishes common proper-motion of the two outer-
most planets and clearly reveals their orbital motion. How-
ever, this baseline is still quite small compared to their orbital
periods (P > 200 yr), and astrometric measurements over a
longer baseline would be highly valuable in better constrain-
ing their orbits, and consequently the star and planets masses.
This will require several years of observations, but could be
achieved more rapidly should a similar re-analysis be applied

Page 2

2Lafrenière et al.
successfully to even earlier data.
HR 8799 was observed with HST/NICMOS in 1998 as part
of a direct imaging survey for massive planets around young
nearby stars (program 7226, PI Eric Becklin); the main re-
sults of this survey were published in Lowrance et al. (2005).
Although these observations have, in principle, sufficient an-
gular resolution and flux sensitivity to see the planets of the
HR 8799 system, the scattered light from the bright primary
prevented their actual detection. This scattered light can be
partly removed by subtracting images obtained at two dif-
ferent spacecraft roll angles, the so-called roll-deconvolution
technique (Schneider & Silverstone 2003), but this is insuffi-
cient to reveal the planets because of important PSF evolution
between the two roll angles. As part of the same HST pro-
gram, many other stars were observed using the same instru-
mental configuration as for HR 8799, and as a result several
images display PSFs very similar to that of HR 8799. As sug-
gested in Lafrenière et al. (2007b), such observations provide
a very interesting basis for constructing an optimized refer-
ence PSF image using the LOCI algorithm. The LOCI al-
gorithm, detailed in Lafrenière et al. (2007b), determines the
coefficients needed to linearly combine several reference im-
ages into an optimized reference PSF image whose subtrac-
tion from the target image will minimize the residual noise.
An important and powerful feature of this algorithm is its
flexibility to optimize the PSF subtraction locally over sev-
eral sub-sections of the image. Applied to the HST data set
mentioned above, the LOCI algorithm could potentially re-
duce the scattered light of HR 8799 sufficiently to reveal the
planets. With this in mind we have re-analyzed these data
and here we report the successful detection of the furthest of
the three planets, HR 8799b, thus extendingastrometric mea-
surements of this planet to ten years. Incidentally, this means
that an exoplanet could have been discovered by direct imag-
ing only a few years after the discovery of 51 Peg b by the
radial velocity technique.
2. DATA SET AND ANALYSIS
All the data used in this study come from HST program
7226, which was a survey for giant planets and brown dwarfs
around young nearby stars. The observations used the NIC-
MOS medium resolution camera NIC2, the F160W filter and
the0.63′′diameteroccultingspottoreducethediffractedlight
from the primary star. Typically three images of each target
were obtained at each of two roll angles differing by 29.9◦.
A total of 39 targets were observed for this program. The
pipeline-reduced images used here were retrieved from the
HST archive housed at the CADC.5
As mentioned above, the LOCI algorithm was used with
this large set of well-correlated images to construct and sub-
tract an optimizedPSF foreach of the six images of HR 8799.
A priori, the set of reference images used by the LOCI algo-
rithmcouldincludeall imagesof targetsthat arenotHR 8799,
but in practice we have omitted some targets from the set be-
cause they showed a second point source within a separation
of3′′fromthestar,weresignificantlyfainterthanHR 8799,or
werenotcenteredbehindthecoronagraph. Thefinalsetofref-
erence images included 203 images from 23 unique sources.
When analyzing a given image of HR 8799, we have added
to this set the three images of HR 8799 obtained at the other
roll angle. Before execution of the LOCI algorithm, all ref-
erence images were registered to the target image based on a
5http://www2.cadc-ccda.hia-iha.nrc-cnrc.gc.ca/cadc/
N
E
N
E
N
E
N
E
ab
cd
FIG. 1.— (a) Image of HR 8799 at first roll angle. (b) Image at the first
roll angle minus image at the second roll angle. (c and d) Residual images,
at both roll angles, after subtraction of an optimized reference PSF image
obtained using the LOCI algorithm, see text for more details. The display is
from 0 to 100 counts s−1for panel a and from -0.5 to 2.5 counts s−1for panels
b–d. Each panel is 4.8′′on a side. The dotted circles in the bottom two panels
mark the inner and outer limits of the optimization annulus used for the LOCI
algorithm. The two short radial lines enclose the position of HR 8799 b.
To produce the images shown in panels c and d, a second optimization was
performed for the half annulus not containing HR 8799 b, such that the full
annulus is representative of the residual noise.
cross-correlation of the secondary mirror support diffraction
spikes. The algorithm was first applied with the specific goal
of detecting HR 8799 b, and thus a single optimization region
was used. This optimization region was defined as a half an-
nulus extending radially from 1.275′′to 2.175′′and centered
on, but excluding pixels within a radius of 5 pixels from the
calculated position of HR 8799 b. The exclusion was used
to prevent the algorithm from trying to subtract the compan-
ion itself, thus biasing the determination of the coefficients
of the linear combination and affecting the residual flux of
the companion. The area of the optimization region, ∼900
pixels, is equivalent to 225 PSF cores (i.e. the characteristic
size of a speckle). This number of “degrees of freedom” in
the optimization region may be compared with the number of
different variables, 46 different observations (23 sources each
at two roll angles), in the linear combination of the optimum
reference image. The different observations do not represent
free parameters of the model PSF, however, as they are highly
correlated with each other.
The companion HR 8799 b was detected in all six resid-
ual images and its position between the two rolls angles is
precisely equal to the spacecraft rotation angle applied. The
threeresidual images at each roll anglewere co-addedandthe
results are shown in Fig. 1. At the peak pixel of the compan-
ion in these co-added images, the detections are at the levels
of 10σ and 6σ for the first and second roll angles, respec-
tively. We have verified that the residual noise in the opti-
mization region closely follows a Gaussian distribution, thus
the significance of the detection is high. The improvement in
PSF subtraction provided by the LOCI algorithm is obvious
from Fig. 1, in which the much higher residuals of the roll-
subtractedimage is apparent. At the separationof HR 8799 b,
the residual noise in the LOCI-subtracted image is nine times
lower than for the classical roll-deconvolutionimage (panel b
of Fig. 1). It is interesting to note that the LOCI subtraction

Page 3

HST/NICMOS detection of HR 8799 b in 19983
TABLE 1
HR 8799 b astrometric measurements
Epoch
∆α (mas)
∆δ (mas) Sep (mas)PA (◦)
1998.8285
2004.5337
2008.6267
1411±9
1471±5
1527±2
986±9
884±5
800±2
1721±12
1716±7
1724±3
55.1±0.4
59.00±0.23
62.35±0.09
NOTE. — Ep. 1998 from this work; others from Marois et al. (2008).
almost reaches the PSF photon noise. At the separation of
HR 8799 b, we estimate that the total residual noise is ∼1.8
times more important than the local photon noise.
To confirm that the above detection is not an artifact from
the image processing done, we have repeated the analysis 1)
without excluding the region containing the companion from
the optimization half-annulus, and 2) using only half of the
available reference images. The companion was still well de-
tected in both cases, although at lower signal-to-noise ratios
(S/Ns) as expected. We have repeated the above analysis with
an optimization region designed specifically for HR 8799 c,
at ∼0.9′′, but the results were inconclusive. The same applies
for d, which is at an even smaller separation.
3. RESULTS
The astrometric and photometricanalysis of the companion
wasdoneusingmodelPSFsgeneratedwiththeTinyTim6soft-
ware (version 7.0)(Krist 1993). Model PSFs appropriate for
the NIC2 pre-cryocooler camera with the F160W filter were
generated for each of three approximate positions on the de-
tector: the central star, the companion at the first roll angle,
and the companion at the second roll angle. All model PSFs
were generated for a width of 10′′and a pixel oversampling
factor of 9 to minimize interpolationeffects when shifting im-
ages. Accordingly,during the analysis all shifts were done on
the oversampled images, which were then binned 9×9 pixels
tomatchthe actualNIC2 platescale. Initially,the modelPSFs
are normalized to a total flux of one.
For each image, the appropriatecompanionmodel PSF was
first spatially shifted and intensity scaled over a grid in dx,
dy, and flux. Then each model PSF in this 3D parameter
space was subtracted from the image and the residual noise in
a 7×7-pixel box centered on the companion was computed.
Thecombinationof dx, dy, andfluxthat yieldedthe minimum
residual noise was found and provided the coordinates and
flux of the companion. The uncertainty on the position of the
companion, 5 mas, was estimated from the dispersion of the
measurements made in all individual images. The center of
thestar was determinedbyshiftingtheappropriatemodelPSF
tomaximizethecross-correlationofits diffractionspikeswith
those of the actual image, as well as by a cross-correlation of
the observed PSF diffraction spikes with themselves after a
rotation of 180◦about the inferred center; the maximum dif-
ferencebetweenthesetwo centroidmeasurements,∼0.1pixel
or 7.5 mas, was taken as an estimate of the centroid accuracy.
The pixel coordinates of the star and companion were con-
verted to RA and DEC using the astrometry solution defined
in the image FITS headers. A S/N-weighed mean of the rel-
ative position of HR 8799 b over the six images was finally
calculated; the results are indicated in Table 1.
Since the model PSFs span 10′′, the companion flux found
by the above procedure is effectively the same as that in
an infinite aperture. The information in the NICMOS Data
6http://www.stsci.edu/software/tinytim/tinytim.html
1.501.45 1.40
∆RA (arcsec)
0.80
0.85
0.90
0.95
1.00
∆DEC (arcsec)
2008.6267
2004.5339
1998.8285
FIG. 2.— Measured relative positions (with associated errors) of HR 8799 b
at the three available epochs. The solid curve shows a Keplerian circular
orbit with a = 71.4 AU, i = 19.2◦, and position angle of ascending node Ω =
131.3◦; the open diamonds mark the corresponding planet position at the
three epochs. The χ2for this orbit is 3.19.
TABLE 2
HR 8799 b photometric measurements
Filter Magnitude
MeasuredIntermediate cloudsExtreme clouds
J
F160W
H
Ks
L′
19.28±0.16
18.54±0.12
17.85±0.17
17.03±0.08
15.64±0.11
19.42
18.58
18.30
17.43
15.24
19.66
18.31
18.08
16.91
15.59
NOTE. — F160W from this work; others from Marois et al. (2008).
Handbook v7.0 was then used to convert this flux (20.5±
2.5countss−1)tophysicalunits,yielding(4.2±0.5)×10−5Jy,
or a magnitude of 18.54±0.12. This magnitude is also in-
dicated in Table 2 along with the measurements reported by
Marois et al. (2008) and the expected magnitudes for two dif-
ferent atmosphere models (see §4 for more detail on models).
4. DISCUSSION
As visible in Fig. 2, the new position is consistent with the
previous ones, and still suggests a nearly circular orbit seen
close to face-on. We have done very simple orbital fits to
verify that the data are consistent with true Keplerian orbits.
Forsimplicity, we haveconsideredonlycircularorbits andas-
sumed that the stellar mass is precisely equal to 1.5 M⊙, and
then we explored a range of semimajor axes (a = 60–100 AU)
and inclinations (i = 0–45◦). The best fits, with a χ2∼ 3.2,
were found for a ∼ 68–74 AU and i ∼ 13–23◦; an example
of such fit is shown in Fig. 2. This range of inclination is in
line with what is expected for the equatorial plane of the star
basedonits measuredvsini (37.5kms−1, Gray & Kaye1999)
which,forthe rangeoftruerotationvelocityofA5 stars (100–

Page 4

4Lafrenière et al.
FIG. 3.— Synthetic spectrum from a 820 K, logg = 4.0 model atmosphere
with vertically stratified clouds. Filled symbols are the band-integrated fluxes
and observations are shown as open symbols with 3σ error bars. Filter re-
sponse curves are plotted below the spectrum.
300 km s−1, Royer et al. 2007), would yield i = 7◦–22◦. We
refrain from carrying out more detailed orbital fits as the lim-
ited time baseline and astrometric precision available would
prevent us from reaching useful constraints on the orbits.
The low luminosity and red near-infrared color of
HR 8799 b (J−Ks= 2.25) are indicative of atmospheric dust,
but constraining the dust properties (composition and distri-
bution) requires a detailed comparison of its SED with var-
ious model atmosphere predictions. Figure 3 compares the
new F160W photometric point, along with earlier J, H, Ks
and L′-band photometry, to an intermediate (vertically strat-
ified) cloud model (Barman et al., in prep.) with parame-
ters selected by comparing the planet’s age and luminosity to
substellar evolution tracks (Marois et al. 2008). The observed
and model photometry are listed in Table 2. The F160W fil-
ter is wide enough to incorporate a substantial fraction of the
water band between the J and H bands. Since the depths of
the water bands are greatly reduced as dust content increases,
this filter, and other NICMOS band-passes, have greater po-
tential for constraining the atmospheric dust content than the
standard ground-based near-IR windows. As can be seen in
Fig. 3 (and Table 2), the new F160W photometric point is
in excellent agreement with this intermediate cloud model,
and about 2σ fainter than an extreme dusty atmosphere hav-
ing clouds that blanket most of the atmosphere. While the
overall agreement with the extreme cloud atmosphere model
appears to be relatively good, based on Table 2, this model
requires Teff= 1600 K (cf ∼800 K for the intermediate cloud
model) and a very small radius (∼4 Earth radii) to match the
observed luminosity. Both of which are inconsistent with for-
mation and evolution models.
5. CONCLUDING REMARKS
The work presented here demonstrates that the LOCI algo-
rithm can take HST data a step further, extending the sensi-
tivity to planets beyond what has been achieved before. The
good performance of the LOCI algorithm with HST data is
due in large part to the high stability of the optical aberra-
tions of HST and the consistent, accurate pointing between
the differentvisits. Over the years, a large archive of HST ob-
servations aimed at detecting exoplanets has been assembled.
Based on the present results, it would be interesting to look
again at this set of data, using the LOCI algorithm, to see if
any other sources have been missed by previous searches. On
a related note, it could be interesting to attempt a re-analysis
of archival ground-based AO observations using an approach
similar to that used here. However, this could turn out be
much less efficient than for HST given the larger evolution of
the PSF structure for ground-basedtelescopes.
Lookingintothe future,the approachpresentedhereshould
definitely be an integral part of the strategy for the search
and characterization of planets with future space telescopes,
such as the James Webb Space Telescope (JWST) or a Ter-
restrial Planet Finder. JWST, for instance, is expected to be
relatively stable in temperature and should not suffer from
the breathing problem experienced by HST. Its PSF should
therefore be more stable than that of HST and the LOCI al-
gorithm will likely performextremely well, especially for ob-
servationsobtainedwithin a givenwavefrontadjustmentcam-
paign, typically every 14 days (Gardner et al. 2006). A pre-
liminary analysis of the JWST PSF temporal evolution done
byMakidon et al.(2008)indeedsuggeststhatPSF subtraction
shouldperformwell. Forthisapproachtowork,enoughatten-
tion should be paid to the accurate positioning of the different
targets, in particular for observations made with the corona-
graphic occulting masks. Having an efficient PSF subtraction
strategy will be absolutely necessary for JWST given its com-
plicated, speckled PSF structure arising from its segmented
pupil, in addition to the speckles arising from the unavoid-
able optical aberrations. While roll-deconvolution could per-
formwell for JWST in the absence of telescope breathing,the
LOCI approach remains very interesting given the observa-
tory’s maximum roll angle of only ±5◦, which will severely
limit the use of roll-deconvolutionat small separations.
The authors would like to thank Yanqin Wu and Andrew
Shannon for discussions of the possible orbits of this system,
BruceMacintoshfordiscussionsrelatingtothenewNICMOS
PSF subtraction approach used here, and Markus Janson for
discussions about various aspects of the NICMOS data. D.L.
and C.M. are supported in part through postdoctoral fellow-
ships from the Fonds Québécois de la Recherche sur la Na-
ture et les Technologies. R.D. is supported through a grant
from the Natural Sciences and EngineeringResearch Council
of Canada.
REFERENCES
Fabrycky, D. C., & Murray-Clay, R. A. 2008, ArXiv e-prints, astro-
ph/0812.0011
Gardner, J. P. et al. 2006, Space Science Reviews, 123, 485
Gray, R. O., & Kaye, A. B. 1999, AJ, 118, 2993
Kalas, P. et al. 2008, Science, 322, 1345
Krist, J. 1993, in ASP Conf. Series, ed. R. J. Hanisch, R. J. V. Brissenden, &
J. Barnes, Vol. 52, 536
Lafrenière, D. et al. 2007a, ApJ, 670, 1367
Lafrenière, D. et al. 2007b, ApJ, 660, 770
Lafrenière, D. et al. 2008, ApJ, 689, L153
Lagrange, A. . et al. 2008, ArXiv e-prints, astro-ph/0705.4290
Lowrance, P. J. et al. 2005, AJ, 130, 1845
Makidon, R. B. et al. 2008, in Proc. SPIE, 7010, 70100O
Marois, C. et al. 2006, ApJ, 641, 556
Marois, C. et al. 2008, Science, 322, 1348
Royer, F., Zorec, J., & Gómez, A. E. 2007, A&A, 463, 671
Schneider, G., & Silverstone, M. D. 2003, in Proc. SPIE, 4860, 1