High-Resolution Transmission Spectroscopy of the Terrestrial Exoplanet GJ 486b: Resultsby@exoplanetology
299 reads

High-Resolution Transmission Spectroscopy of the Terrestrial Exoplanet GJ 486b: Results

tldt arrow

Too Long; Didn't Read

The exoplanet GJ 486b, orbiting an M3.5 star, is expected to have one of the strongest transmission spectroscopy signals among known terrestrial exoplanets.
featured image - High-Resolution Transmission Spectroscopy of the Terrestrial Exoplanet GJ 486b: Results
Exoplanetology.Tech: Research on the Study of Planets HackerNoon profile picture

This paper is available on arxiv under CC 4.0 license.


(1) Andrew Ridden-Harper, Department of Astronomy and Carl Sagan Institute, Cornell University & Las Cumbres Observatory;

(2) Stevanus K. Nugroho, Astrobiology Center & Japan & National Astronomical Observatory of Japan;

(3) Laura Flagg, Department of Astronomy and Carl Sagan Institute, Cornell University;

(4) Ray Jayawardhana, Department of Astronomy, Cornell University;

(5) Jake D. Turner, Department of Astronomy and Carl Sagan Institute, Cornell University & NHFP Sagan Fellow;

(6) Ernst de Mooij, Astrophysics Research Centre, School of Mathematics and Physics & Queen’s University Belfast;

(7) Ryan MacDonald, Department of Astronomy and Carl Sagan Institute;

(8) Emily Deibert, David A. Dunlap Department of Astronomy & Astrophysics, University of Toronto & Gemini Observatory, NSF’s NOIRLab;

(9) Motohide Tamura, Dunlap Institute for Astronomy & Astrophysics, University of Toronto

(10) Takayuki Kotani, Department of Astronomy, Graduate School of Science, The University of Tokyo, Astrobiology Center & National Astronomical Observatory of Japan;

(11) Teruyuki Hirano, Astrobiology Center, National Astronomical Observatory of Japan & Department of Astronomical Science, The Graduate University for Advanced Studies;

(12) Masayuki Kuzuhara, Las Cumbres Observatory & Astrobiology Center;

(13) Masashi Omiya, Las Cumbres Observatory & Astrobiology Center;

(14) Nobuhiko Kusakabe, Las Cumbres Observatory & Astrobiology Center.


We cross-correlated the models described in Section 4 with the processed data and searched for peaks in the phase-folded cross-correlation function at the expected KP and vsys of GJ 486b. Figs. 2 − 8 show the results of our search for H2O, CO2, HCN, NH3, CH4, Na, and the 700 K solar abundance model with condensation, respectively. These figures show the 1D cross-correlation functions phase-folded using GJ 486b’s expected value of KP as well as its negative value. We show the result of using −KP as we later use it to inject model spectra into the data in Section 6.2.

We find no evidence for robust correlations between any of our models[9] and the data. The most promising feature is of H2O in the IGRINS data set (see Fig. 2), which is at the 3σ level[10] in the expected position in KP and vsys. However, it is likely that this feature is produced by imperfectly removed water lines in the spectrum of GJ 486b’s M-dwarf host star, given that it is present after applying only one iteration of SYSREM, as shown in Fig. 9.

We do not understand why these residuals were not removed by SYSREM and only appear in the IGRINS data set. However, we note that they appear to be qualitatively similar to those reported by Brogi et al. (2013). In their case, Brogi et al. (2013) detected molecular absorption in the dayside emission spectrum of 51 Pegasi b on two separate occasions, but not a third occasion when strong stellar residuals were present in the data. Chiavassa & Brogi (2019) revisited this data and used 3D simulations of stellar convection to remove the stellar spectrum. This method better suppressed the stellar residuals in the third observation and allowed the planetary signal to be detected. In a future study, it may be beneficial to investigate whether the method of Chiavassa & Brogi (2019) can better suppress the stellar residuals in our IGRINS data.

Regarding our nondetection of Na, Caballero et al. (2022) suggest that if GJ 486b were to have a tenuous atmosphere with a surface pressure of 10−6 bar, nonthermal processes such as surface sputtering by ions may heat its atmosphere to temperatures on the order of 10,000 K. They show that a Na abundance of 10−9 in such an atmosphere would produce transit depths of approximately 1% in the cores of the Na D lines (5889.95 and 5895.92 ˚A), which could be detected by observing only a few transits with the Calar Alto high-Resolution search for M dwarfs with Exoearths with Near-infrared and optical Echelle Spectrographs (CARMENES; Quirrenbach et al. 2014). The spectral ranges of our observations do not include the Na D lines, but our IRD and SPIRou data sets do include the Na lines at 1.138 and 1.141 µm, which are much weaker and more difficult to detect. Therefore, while we found no evidence for a nonthermal Na atmosphere, we do not consider this to be evidence for its absence. We recommend further investigation of this scenario with additional observations targeting the Na D lines.

6.2. Constraining GJ 486b’s Atmospheric Composition

As we did not detect the atmosphere of GJ 486b, here we determine the abundances that our data can rule out under the assumption that GJ 486b’s atmosphere is clear. We do this by injecting cloud-free model spectra into the reduced data before processing and crosscorrelating as in Sections 3 and 5. For each in-transit frame, we injected the model transmission spectrum at the velocity given by Equation 2 if KP were replaced by its negative value. We injected the signals at −KP so that they would not be affected by any weak real planet signals that may be present in the data.

To optimally combine the different data sets, we summed the log-likelihoods. Our cross-correlation to log-likelihood mapping produced an array with dimensions KP, vsys, and α. Examples of the recovered conditional (2D) and marginalized (1D) likelihood distributions after injecting a water model into the data are shown in Fig. 10.

We used the marginalized distributions of KP, vsys, and α (as shown in Fig. 10) to find their most likely values and their associated 1σ errors. With this approach, the significance to which an injected signal was recovered is well approximated by how well α is constrained (Gibson et al. 2020, 2022). This method works because we assume that marginalized distributions of α that are constant at α = 0, or unconstrained, correspond to a nondetection. Therefore, we are essentially measuring how significantly α differs from zero.

Figure 2. The results of cross-correlating the SPIRou (first row), IRD (second row), and IGRINS (third row) data sets with a model transmission spectrum containing spectral lines only from H2O. The bottom row shows the result of combining all data sets together. The first column shows the cross-correlation for each frame in the barycentric frame. The expected trace of the planet’s signal is shown as the dotted diagonal line. The second column shows the cross-correlation for each frame in the planet’s rest frame. The dotted line shows the expected trace of the planet’s signal at the system’s systemic velocity. The third column shows the phase-folded cross-correlation signal as a function of phase. The dotted horizontal and vertical lines show the planet’s expected position in ±KP and systemic velocity, respectively. The fourth column shows the 1D cross correlation as a function of systemic velocity at ±KP. The black dotted line indicates the system’s systemic velocity.

Figure 3. Same as Fig. 2 except for CO2.

However, in practice, our recovered values of α were slightly less than the true values, leading to a slight underestimate of the true recovered detection significances.

Figure 4. Same as Fig. 2 except for HCN. 0.028

Figure 5. Same as Fig. 2 except for NH3.

This is because we did not explicitly account for how SYSREM distorts an underlying planet signal when removing the stellar and telluric lines. Gibson et al. (2022) showed that this can be corrected by pre-processing the models in an analogous way to how SYSREM processes the planet signal in the data. We investigated the feasibility of using this approach by trying it for a single model spectrum and a typical grid covering ±60 km s−1 around the expected KP and vsys with step sizes of 1 km s −1 . The run time of this trial on a reasonably capable workstation[11] was 80 minutes. Our full analysis uses about 640 different model spectra, which would therefore require a total run time of about 36 days. We consider this to be infeasible for this study, so we instead report our upper limits with the caveat that they likely underestimate the true upper limits.

Figure 6. Same as Fig. 2 except for CH4.

Figure 7. Same as Fig. 2 except for Na. The row for IGRINS is omitted because no Na lines are within its spectral range.

We were able to recover some injected models for H2O, CH4, NH3, HCN, CO2, and CO, indicating that our high-resolution observations can constrain the abundances of these chemical species under the assumption of GJ 486b having a clear atmosphere. As these species have different molecular weights, line strengths, and numbers of lines, they are each sensitive to different regions of VMR-MMW parameter space. These constraints are shown in Fig 11 for H2O, CH4, NH3, HCN, CO2, and CO, respectively. Each figure shows the VMRMMW constraints for the individual data sets (SPIRou, IRD, IGRINS) and all data sets combined. Models with higher MMWs have smaller atmospheric scale heights, making them more difficult to recover. Injected models that were recovered to significances of ≥5σ and 3≤σ<5 are colored in black and gray, respectively. Injected models that were recovered to significances of 0≤σ<3 are considered to have not been recovered and are shown in light blue. This means that we can rule out model atmospheres to 5σ and 3σ in the VMR-MMW regions colored black and gray, respectively. Conversely, the light blue regions represent atmospheric compositions that are allowed by our data.

Figure 8. Same as Fig. 2 except for using the solar abundances at 700 K model generated by GGchem.

We obtain strong constraints on the presence of H2O in GJ 486b’s atmosphere. Assuming a clear atmosphere, our observations rule out log10(VMR) ≥ −3 for atmospheric MMWs ≤ 5 and log10(VMR) ≥ −4 for MMWs ≤2.5 to 5σ. Therefore, our observations are inconsistent with a clear H2/He-dominated atmosphere with solar (or somewhat subsolar) H2O abundances. We can also, less confidently, rule out a clear atmosphere of log10(VMR) ≥ −2 for MMWs ≤ 16 and a clear 100% H2O atmosphere to 3σ. Our results, therefore, suggest that GJ 486b does not have a clear H2O-dominated atmosphere. This could have important implications for planetary interior models, which often predict H2O-dominated atmospheres as the most likely possibility for 3 M⊕ planets like GJ 486b (e.g., Ortenzi et al. 2020).

Our derived constraints on the abundances of CH4, NH3, HCN, CO2, and CO are summarized in Table 2. These constraints show that our observations are inconsistent with a clear H2/He-dominated atmosphere with a solar CH4 abundance. However, solar abundances of NH3, HCN, CO2, and CO are consistent with our data. For cloud-free models including all the above species at solar abundances, we rule out to 5σ those with temperatures > 500 K, as shown in Fig. 12.

We were not able to recover any of our injected models for C2H2, FeH, H2S, Na, K, PH3, SiO, TiO, or VO. This indicates that, regardless of their VMR or the background atmosphere MMW, our observations are not sensitive to these chemical species.

For all the species considered here, the data from IGRINS on Gemini-S offers the strongest limits. This likely results from Gemini-S having a larger collecting area than SPIRou’s CFHT and the wavelength coverage (1.45-2.45 µm) of IGRINS extending further into the infrared than that of IRD (0.97-1.75 µm).

Overall, these upper limits indicate that GJ 486b lacks a clear H/He-dominated atmosphere with solar abundances. However, some possible cloudy atmospheres or high-MMW secondary atmospheres are allowed by our data.

[9] Including both the Exomol (Yurchenko & Tennyson 2014) and HITEMP 2020 (Hargreaves et al. 2020) opacity data for CH4.

[10] Approximated as the amplitude of the cross-correlation function at the expected velocity divided by the standard deviation of the surrounding regions.

[11] With the following specifications: Ryzen 5900X (12 core) CPU, 128 GB of DDR 3200 RAM, and a PCIe 4.0 M.2 SSD.