Detection of Carbon Monoxide and Water Absorption Lines in an Exoplanet Atmosphere

See allHide authors and affiliations

Science  22 Mar 2013:
Vol. 339, Issue 6126, pp. 1398-1401
DOI: 10.1126/science.1232003

High-Resolution Spectrum of an Exoplanet

Unlike most of the extrasolar planets we know about, the four planets around the star HR 8799 were detected directly. Konopacky et al. (p. 1398, published online 14 March; see the Perspective by Marley) obtained a high-resolution spectrum of one of the planets that reveals both water and carbon monoxide but not methane in the planet's atmosphere. The atmospheric carbon-to-oxygen ratio, which traces the process of planet formation, is greater than that of the host star, providing clues to how the planets formed.


Determining the atmospheric structure and chemical composition of an exoplanet remains a formidable goal. Fortunately, advancements in the study of exoplanets and their atmospheres have come in the form of direct imaging—spatially resolving the planet from its parent star—which enables high-resolution spectroscopy of self-luminous planets in jovian-like orbits. Here, we present a spectrum with numerous, well-resolved molecular lines from both water and carbon monoxide from a massive planet orbiting less than 40 astronomical units from the star HR 8799. These data reveal the planet’s chemical composition, atmospheric structure, and surface gravity, confirming that it is indeed a young planet. The spectral lines suggest an atmospheric carbon-to-oxygen ratio that is greater than that of the host star, providing hints about the planet’s formation.

HR 8799 is a young [~30 million years (My)] (1) early type (A5 to F0) star ~130 light years from the Sun. Near-infrared direct imaging with adaptive optics revealed four planets around this star (2, 3). A number of studies have broadly characterized these planets, showing that all four have masses between 5 and 10 times that of Jupiter (MJup), effective temperatures between 900 and 1200 K (still hot from the gravitational energy released during their formation), and near-infrared colors that are redder than initially expected for their mass and temperature ranges. The red colors are best explained by the presence of iron and silicate atmospheric clouds. These clouds are normally located below the photosphere in old field brown dwarfs with temperatures less than 1400 K, but they are believed to persist at cooler temperatures in young planets where surface gravities are much lower (46). Low surface gravity may also contribute to extreme deviations from equilibrium concentrations of CO and CH4, potentially explaining the lack of strong CH4 absorption (7) previously anticipated in young planets. The formation mechanism of the HR 8799 planets—and all exoplanets—remains uncertain, with both global disk instabilities (8) and bottom-up core accretion (9, 10) being proposed. One possible way to differentiate between these is through planetary composition. Planets that form through disk instabilities will track the bulk abundance of the original star-forming material, whereas core-accretion planets may be enhanced or depleted in some elements.

Further inferences on the chemical composition and surface gravity of the HR 8799 planets, and exoplanets in general, have thus far been limited by poor spectral resolution (typically λ/Δλ ~ 10 to 100; λ, wavelength). Here, we present a moderate-resolution (λ/Δλ ~ 4000) spectrum of the planet HR 8799c. We obtained our observations of HR 8799c in 2010 and 2011 with the Keck II 10-m telescope, using the integral field spectrograph OSIRIS (OH-Suppressing Infrared Integral Field Spectrograph) (11) in conjunction with the facility adaptive optics system (12). These observations were obtained in the K band (1.965 to 2.381 μm) at the finest spatial sampling provided by OSIRIS (0.02′′ per spaxel). A total of 5.5 hours of observations were obtained, consisting of 33 exposures, each 600 s long. We processed the raw data through sky and dark subtraction, bad pixel removal, and telluric calibration from an A0 stellar standard, and we then transformed these data into data cubes (13). Each data cube has ~1600 spectral channels with a 0.32′′-by-1.28′′ field of view at a nominal spectral resolution (λ/Δλ) ~4000 per spatial location.

Scattered starlight in the form of an interference pattern of speckles makes identification of the planet challenging. We used speckle modeling and subtraction (14) for spectral extraction. A key insight is that the residual speckle noise has a characteristic spectral signature: generally broad (500 Å) features passing through the planetary spectrum. This produces highly correlated noise in a low-resolution spectrum over the full band pass, but does not affect the detectability of individual narrow spectral lines. We extracted the spectra from each of the 33 data cubes and combined them into a single spectrum with an average signal-to-noise ratio per channel of ~30 (15).

Only a handful of objects with mass and age similar to HR 8799c are known, and broad K-band spectra are available for only a few of these, all with resolutions substantially lower than 4000. We compared the extracted spectrum of HR 8799c to the spectrum of a ~5 MJup brown dwarf, 2MASS J12073346-3932539b (2M1207b) (16), and the spectrum of the planet HR 8799b (14), all three binned to the same resolution (Fig. 1). HR 8799c falls between 2M1207b and HR 8799b in appearance and, like 2M1207b, has an obvious CO bandhead at ~2.29 μm. The spectral morphology and colors of HR 8799b and 2M1207b also imply the presence of dust clouds in their atmospheres (14, 17, 18). The rough morphological similarity between HR 8799c and these objects implies that HR 8799c also has a cloudy atmosphere. The inference of atmospheric clouds is consistent with results from studies based on the photometry of HR 8799c (2, 1721), which find that thick dust clouds are necessary to explain its spectral energy distribution.

Fig. 1

Low-resolution (λ/Δλ ~ 50) binned K-band spectrum for HR 8799c compared with the spectra of HR 8799b and 2M1207b. Binning the data bypasses issues related to the different resolutions and sampling among the three data sets while allowing us to compare broad molecular absorption bands (note that the residual speckle noise is correlated between the individual spectral channels, though this does not affect the analysis here). Each spectrum shows expected broad H2O absorption at λ < 2.15 μm that appears strongest in HR 8799b. Whereas CH4 and CO absorption probably contribute to the downward slope redward of ~2.15 μm for HR 8799b (14, 40), their absorption is substantially weaker than is typical for an object near 1100 K. The spectrum of 2M1207b, however, has no discernible contribution from CH4 and is likely shaped primarily by clouds, H2O, and CO opacity (7). The best-fitting model spectrum for HR 8799c, with a temperature of 1100 K and a surface gravity of 100 m s−2 [log(g) = 4.0, cgs units], is shown in green. The best-fitting temperatures and gravities for HR 8799b and 2M1207b are ~750 to 1100 K and log(g) ~ 3.5 to 4.5 (17) and ~1000 K and log(g) ~ 4.0 (7), respectively. Fλ, flux in units of wavelength.

With the speckles mostly removed, the long-exposure (5.5 hours) data are sufficient to examine the spectrum at the nominal resolution of OSIRIS (Fig. 2). Numerous spectral lines are immediately obvious, particularly those corresponding to strong CO (2,0) and (3,1) band heads at 2.29 and 2.32 μm. The remaining absorption lines are expected to be due to H2O.

Fig. 2

K-band spectrum of HR 8799c (black) at the full resolution of OSIRIS (λ/Δλ ~ 4000). Both lines and continuum are shown, and the spectral features most relevant to objects of this mass are highlighted. For clarity, uncertainties are shown separately in red. The best-fitting model spectrum is shown in green. A clear drop from CO is detected, along with features typical of the CO bandhead at λ > 2.29 μm. The slight increase in the spectrum at the red end is attributed to residual speckle effects after the attenuation algorithm. Again, broad features from CH4 are not easily detected. Spectral data are provided in database S1 (15).

To further verify that the features are, in fact, molecular lines and not residual noise, the spectrum was low-pass filtered. Filtering effectively removes any residual speckle artifacts in the spectrum, as well as the planet’s continuum, leaving behind only the absorption lines (Fig. 3) (15). Because the scattered light has roughly the spectrum of HR 8799A, a nearly featureless A-type star, modulated at 500 Å scales by the speckle process, all remaining narrow spectral lines are from the planet. Although many CO and H2O lines are visible, we did not identify any CH4 lines. We cross-correlated the low-pass–filtered planetary spectrum with spectral templates of pure H2O, CO, and CH4. Given the incompleteness of methane molecular line data, we used three popular, yet fairly distinct CH4 lists as templates (2225) (Fig. 3). The cross-correlation confirms that all lines detected in the low-pass–filtered observations are attributable to CO and H2O.

Fig. 3

(Top) Pure H2O (blue), CO (green), and CH4 (orange) synthetic spectra demonstrate the predicted location of absorption lines (14). The filtered spectrum of HR 8799c is shown in red. A filtered model atmosphere spectrum of mostly H2O and CO is overplotted in black. Also shown is a spectrum of a bright speckle (black trace at bottom), scaled such that the variance is equal to the variance in a featureless region of HR 8799c. (Bottom) Cross-correlation functions for the spectrum of HR 8799c and the synthetic spectra shown in the top panel (solid curves), along with a baseline cross-correlation between the planet spectrum and a bright speckle (dotted curve). The cross-correlation with the H2O-only template covered the entire observed wavelength range. Correlations with the CO and CH4 templates were performed only over wavelength regions with strong lines (CH4: λ > 2.18 μm; CO: λ > 2.29 μm). Large cross-correlation peaks are found for the pure CO and H2O templates, as expected. The CO-only template produces two smaller symmetric peaks at ± 207 km s−1; this velocity corresponds to the near-equal line spacing of the CO lines across the (2,0) bandhead starting at 2.29 μm. Similar near-equal line spacing occurs for other CO bandheads [that is, the (3,1) bandhead at 2.32 μm], resulting in the ringing behavior seen in the cross-correlation. No peaks in the correlations were found with any of the three CH4 templates or the speckle spectrum (this is true for any subset of the observed wavelength range used in this work). Cross-correlation is not required to detect molecular lines in our spectrum; however, this exercise aids in quantifying the relative detections (or nondetection, in the case of CH4) when lines from individual molecules overlap.

Two quantities that greatly influence a planet’s atmosphere are effective temperature and gravity. The effective temperature of HR8799c is likely between 900 and 1300 K based on its observed bolometric luminosity (L) [–4.8 < log(L/L) < –4.6, where L is the luminosity of the Sun] (2) and the range of radii predicted by formation and evolution models (1 to 1.5 Jupiter radii). Dynamical stability requirements of the planetary system require all four planets to be less than 10 MJup (2, 26). This mass limit, combined with the above radius range, implies a surface gravity below 250 m s–2. Furthermore, the planetary system’s youth (~30 My) (1, 27) and cooling tracks for giant planets indicate a surface gravity between ~30 and ~250 m s−2.

The upper limit we set on rotation (vsini; where v is velocity and i is inclination of the rotation axis) for HR 8799c is ~40 km s−1, which is roughly equal to the resolution of our data, indicating that the cumulative effect of broadening by gravity and rotation is modest. We compared the grid of model spectra from (14), convolved to the nominal OSIRIS resolution (~5 Å) and low-pass filtered in the same manner as the observed spectrum, to the drop in flux across the CO (2,0) bandhead, a feature known to be sensitive to gravity (15). Surface gravities of ~100 m s−2 or less, consistent with the range based on the system’s youth and dynamical stability, provide the best match.

We further refined the effective temperature using the K-band spectrum and the surface gravity limit (<100 m s−2). The K-band spectrum has uniformly characterized systematics and uncertainties and contains key features: the overall shape and the CO bandhead depth. Fitting the binned, low-resolution spectrum with a subset of the model grid with log(g) = 3.5 to 4.0 [centimeter-gram-second (cgs) units; g, surface gravity] results in effective temperature (Teff) = 1100 ± 100 K (15). The best-matching model (Figs. 1 and 2) produces a cross-correlation peak that exceeds the correlations described above (Fig. 3). The model that best matches our K-band spectrum also compares well to the available photometry, including recent photometry (18), with χ2 values equal to or smaller than previously published fits (2, 1721). Previous results have converged on this temperature and gravity range by fitting photometric observations with broad wavelength coverage (17). For the remainder of the analysis, Teff = 1100 ± 100 K and log(g) = 3.5 to 4.0 (cgs units) are adopted as the most plausible values. The implied mass and age ranges for these values are ~3 to 7 MJup and ~3 to 30 My (hot start models) (28).

Our data require the mixing ratio of CH4 to be less than ~10–5 (assuming the methane templates are accurate for the strongest lines) (25). At Teff ~ 1100 K, such a low CH4 mixing ratio indicates that vertical mixing has quenched the CO and CH4 mole fractions at depths corresponding to their maximum and minimum values, respectively. Though many details concerning vertical mixing of CO and its effect on chemical equilibrium are poorly understood, it is well known that reestablishing equilibrium CO/CH4 after CO has been mixed into a normally CH4-rich photosphere can take 105 to 109 years (29). Therefore, even modest vertical mixing will quench the CO and CH4 (and H2O) in the atmosphere where the chemical time scales exceed the mixing time scales (30). Deep quenching requires short mixing time scales (tmix ~ Hp2/Kzz, where Hp is the pressure scale height and Kzz is the eddy diffusion coefficient), which justify the use of a large Kzz = 108 cm2 s−1 in our models.

Of course, the final quenched mixing ratios of CO, CH4, and H2O all depend on the bulk atmospheric elemental abundances that, so far, have been assumed to be equal to that of the Sun (31, 32). The planetary atmospheric abundances may differ from those of the host star, depending on how the planet formed, adding additional importance to chemical composition.

Under gravitational instability (GI), planets are the product of disk instabilities that gravitationally collapse, and though there are several important stages of the collapse, the planet interior and atmosphere are formed simultaneously with a stellar composition (ultimately, that of the molecular cloud from which the system formed). Deviations from a stellar composition are possible through postformation capture of solid material; however, if the HR 8799 planets are massive (>3 MJup), the time scale for substantial planetesimal capture is too short (especially at their current separations) to allow for compositions very different from stellar (33). If the HR8799 planets formed by GI, the spectrum of HR8799c should indicate a stellar composition and, in particular, a stellar C/O ratio (assumed to be equal to that of the Sun for HR8799) (34).

Under core accretion (CA), planets form in a multistep process involving the initial formation of a core (on the order of 10 Earth-masses of heavy elements) followed by runaway accretion, primarily of gas supplied by the disk. When the disk is no longer able to supply a substantial amount of material, the newly formed planet is isolated from what remains of the disk. The final atmospheric composition of a planet formed by CA depends on its location within the disk and the contribution of solids during the runaway accretion phase. A variety of compositions are possible (including stellar C/O with sufficient solid accretion), but a nonstellar composition is highly likely for massive giant planets.

Within the disk that formed the HR8799 planets, there are three important boundaries: the H2O [~10 astronomical units (AU)], CO2 (~90 AU), and CO (~600 AU) frost lines. The planets currently orbit their star between 15 and 70 AU; therefore, are all located between the H2O and CO2 lines where the gas-phase C and O abundances in the disk would have been reduced through the formation of ice and carbon and silicate grains (with CO and CO2 remaining in the gas phase). Therefore, in the CA scenario, planetary atmospheres acquired through gas-only accretion will have substellar C and O abundances but superstellar C/O ratios, because water ice is more abundant than carbon-bearing grains. A simple model of ice formation suggests that the disk gas-phase C/O ratio ~0.9 (35). Increasing the fraction of the atmosphere acquired by solid accretion can lead to superstellar values of both C and O, with the C abundance rising more slowly than that of O, and an overall decrease in C/O (35). Between the CO2 and CO ice-lines, the abundances follow a similar pattern.

To explore the consequences of nonstellar C and O abundances, we made a grid of planetary atmosphere models following that of Barman et al. (14) but using the C and O values predicted by (35) for different ratios of solid to gas accretion, with C/O ranging from 0.45 to 1 (tables S1 and S2). Again, we assumed that the C and O abundances of HR 8799A are solar (15, 32, 34). We found the best fit from this grid of model atmosphere spectra by minimizing χ2 (Fig. 4). Though a comprehensive range of C and O values, independent of any disk-chemistry model, has not been explored, the results from these fits suggest that the C/O ratio is certainly less than 1 and not substellar, but is probably larger than the solar/stellar ratio with substellar C and O.

Fig. 4

Comparison of synthetic spectra (red) with different carbon and oxygen abundances to the low-pass–filtered spectrum of HR8799c (black). The central wavelength range is omitted due to the lack of strong absorption features. Under the assumption that HR 8799c formed near its current location (i.e., little to no migration), the best-matching carbon and oxygen abundances from the model grid are 8.33 and 8.51, respectively (~0.06 and 0.13 below the solar values, using the traditional astronomical log base-10 abundance scale in which the hydrogen abundance is 12.0), corresponding to a C/O ratio ~ 0.65 (middle panel). If HR 8799c somehow formed beyond 100 AU and migrated to its current location, the best-matching C and O abundances are both substellar: 8.29 and 8.45, respectively, with C/O ~ 0.7 (not depicted). In either case, large C/O (~0.9, bottom panel) results in a substantial increase in CH4 and a decrease in H2O in the spectrum, whereas for small C/O (~0.44, top panel), CO and H2O increase. Thus, both cases increase or decrease the prominent molecular lines identified in the K-band spectrum by factors that are easily excluded by the data (for instance, the large CH4 absorption feature at 2.32 μm for C/O ~0.9). Note that the exclusion of the high and low C/O values is independent of Kzz because the adopted value (108) always quenches the CO and CH4 mole fractions at their maximum and minimum values, respectively. Raising Kzz would have no effect, and lowering Kzz would only increase CH4. All abundance values for C and O in the model grid are given in tables S1 and S2.

Although it is fairly straightforward to rule out the extrema of C/O ratios, understanding the uncertainty in our C/O measurement is less so. In addition to noise in the data, the uncertainties in temperature and surface gravity also contribute to measurement errors in C and O, as line depths are sensitive to both of these bulk parameters. To marginalize over the temperature and gravity uncertainties, we expanded the grid of atmosphere models (with abundances given in table S1) to include temperatures of 1000 to 1200 K and log(g) of 3.5 to 4.0. We then performed a Monte Carlo simulation by resampling the spectral data from Gaussian distributions with widths determined by the (uncorrelated) uncertainties for all wavelength channels. We fit each resampled spectrum using the model grid and recorded the best-fit abundances. The resulting estimate (and uncertainty) for the C/O ratio is 0.650.05+0.10, which is marginally greater than the assumed stellar ratio (~0.55).

Measuring abundances is complicated, and model atmospheres have not been thoroughly tested for systematic under- or overestimation of C/O ratios in substellar objects. However, given the dominance of H2O, CO, and CH4 opacities in their atmospheres, large variations in C/O should be easier to discern in brown dwarfs and giant planets than in stars (36). Planet migration and chemical evolution within the disk can muddy conclusions based on composition (37), as can core dredging, which may be important in planets more massive than Jupiter (38). With these caveats in mind, the above analysis rules out a planetary atmosphere for HR8799c that formed by gas-only accretion during a CA process at its current location (C/O > 0.9) and marginally excludes an atmosphere that formed from extreme amounts of solid accretion (C/O < 0.6). Between these extreme predictions, the picture is more complicated, but the enhanced C/O ratio and the depleted C and O levels tend to favor a history in which the planet formed via CA. In this case, after an initial solid core formed, the planet atmosphere accreted from material that was partially, but not completely depleted of solid planetesimals. However, given the uncertainties in parameters such as the stellar abundances (15), we cannot totally exclude a GI formation scenario for this system. Our work shows the power of high-resolution spectra, which allow molecular species to be seen directly through their individual absorption lines rather than inferred from overall spectral shapes that are more sensitive to model parameters.

Supplementary Materials

Materials and Methods

Figs. S1 to S4

Tables S1 and S2

References (4144)

Database S1

References and Notes

  1. See the supplementary materials on Science Online for more information.
  2. The spectrum of HR 8799, a classic Lambda Bootes star, shows substantial depletion of some elements. However, its peculiar abundances have been attributed to its extremely shallow photosphere and reaccretion of gas depleted in refractory elements—possibly associated with planet formation (39)—so we assume here that the planets formed from solar-abundance material.
  3. Acknowledgments: We thank A. Conrad, S. Dahm, J. Lyke, H. Tran, H. Hershley, J. McIlroy, J. Rivera, and the entire Keck staff for maximizing our observing efficiency; J. Larkin and S. Wright for their assistance with OSIRIS data reduction; and R. Murray-Clay and K. Öberg for assistance with abundance values. Portions of this work were performed under the auspices of the U.S. Department of Energy by LLNL under contract no. DE-AC52-07NA27344. Support for this work was provided by NASA Origins of the Solar System grants to LLNL and Lowell Observatory and from the Keck Principal Investigator Data Analysis Fund, managed by NExScI on behalf of NASA. Support was also provided by the NASA High-End Computing Program through the NASA Advanced Supercomputing Division at Ames Research Center. Q.M.K. is a Dunlap Fellow at the Dunlap Institute for Astronomy and Astrophysics, University of Toronto. The Dunlap Institute is funded through an endowment established by the David Dunlap family and the University of Toronto. The W. M. Keck Observatory is operated as a scientific partnership among the California Institute of Technology, the University of California, and NASA. The Keck Observatory was made possible by the generous financial support of the W. M. Keck Foundation. We also wish to recognize the very important cultural role and reverence that the summit of Mauna Kea has always had within the indigenous Hawaiian community. We are most fortunate to have the opportunity to conduct observations from this mountain. All authors contributed equally to this work.
View Abstract

Navigate This Article