The mid-infrared spectrum of $β$ Pictoris b. First VLTI/MATISSE interferometric observations of an exoplanet

TL;DR

This paper reports the first mid-infrared exoplanet spectrum obtained with VLTI/MATISSE using the GRA4MAT fringe-tracking upgrade, targeting β Pictoris b in $L$ and $M$ bands at $R=500$. Through a custom phase-correction pipeline and forward modelling with ForMoSA/Exo-REM, the authors derive a solar-like C/O ratio ($C/O \approx 0.539$) and atmospheric parameters of $T_ ext{eff} \approx 1530$ K, $\log g \approx 3.83$, [M/H] $\approx 0.30$, with a mass around $11\,M_ ext{Jup}$ once anchored to dynamical constraints. The spectrum shows broad $\mathrm{H_2O}$ and $CO$ absorption features in the $L$ and $M$ bands, respectively, validating mid-infrared interferometry as a viable route to atmospheric characterization at small angular separations and highlighting the potential of LM-band observations to complement JWST. The results demonstrate high signal-to-noise and establish a methodology to extract planet spectra in the mid-infrared, paving the way for future observations of closer-in or fainter companions as AO upgrades (GPAO) and Gaia-driven targets expand the sample.

Abstract

Few spectra of directly-imaged exoplanets have been obtained in the mid-infrared (> 3 $μ$m). This region is particularly rich in molecular spectral signatures, whose measurements can help recover atmospheric parameters and provide a better understanding of giant planet formation and atmospheric dynamics. In the past years, exoplanet interferometry with the VLTI/GRAVITY instrument has provided medium-resolution spectra of a dozen substellar companions in the near infrared. The 100-meter interferometric baselines allow for the stellar and planetary signals to be efficiently disentangled at close angular separations (< 0.3''). We aim to extend this technique to the mid-infrared using MATISSE, the VLTI's mid-infrared spectro-interferometer. We take advantage of the fringe tracking and off-axis pointing capabilities recently brought by the GRA4MAT upgrade. Using this new mode, we observed the giant planet $β$ Pictoris b in L and M bands (2.75-5 $μ$m) at a spectral resolution of 500. We developed a method to correct chromatic dispersion and non-common paths effects in the fringe phase and modelled the planet astrometry and stellar contamination. We obtained a high-signal-to-noise spectrum of $β$ Pictoris b, showing the planet continuum in L (for the first time) and M bands, which contains broad absorption features of H$_2$O and CO. In conjunction with a new GRAVITY spectrum, we modelled it with the ForMoSA nested sampling tool and the Exo-REM grid of atmospheric models, and found a solar carbon-to-oxygen ratio in the planet atmosphere. This study opens the way to the characterization of fainter and closer-in planets with MATISSE, which could complement the JWST at angular separations too close for it to obtain exoplanet spectra. Starting in 2025, the new adaptive optics system brought by the GRAVITY+ upgrade will further extend the detection limits of MATISSE.

Figures (17)

• Figure 1: Simulation of the gain in contrast achieved by offsetting the pinhole of MATISSE on $\beta$ Pic b, separated here by 534 mas from its host star. The upper and lower plots depict the stellar and planetary contribution to the coherent flux (the fringe envelope) as seen through the circular MATISSE pinhole (white area, hence the Airy-like shape of the output flux) centred on the star and on the planet, respectively. They are simulated at a wavelength of 3.5 µm and normalized to the maximal intensity of the star-centred stellar PSF. When the pinhole is centred on the planet, the speckle flux is reduced to the same level as the planetary flux. In each simulation, the asymmetric shape of the PSF of the offset object (relative to the pinhole) is due to imperfect spatial filtering of the Airy rings falling in the pinhole.
• Figure 2: Measured differential phase (grey dots) in two MATISSE frames, one on the star (top) and the other on the planet (bottom). Overlaid are the models presented in this work using the air refractive indices of Voronin2017 (blue), Mathar2007 (green), and Ciddor1996 (red). In the absence of chromatic or non-common path OPD, the differential phase should be zero on the star or an oscillation on the planet.
• Figure 3: Real (left) and imaginary (right) parts of the measured planet-to-star coherent flux ratio, $\mathcal{F}$, on the six UT baselines, during one 10 s frame. The error range is plotted in the background. The thick lines show the best-fit model as described in Sect. \ref{['sec:astrometry']}. It includes the star-planet modulation, the stellar speckle contamination (dashed blue curves), and the planet-to-star contrast assumption (dashed black curve).
• Figure 4: Scaling of the BT-Nextgen stellar model ($T_{\mathrm{eff}} = 7890$ K, $\log g = 3.83$, [Fe/H] = 0.0) with species using Tycho, Gaia, and 2MASS photometry of $\beta$ Pictoris. The bottom panel shows the residuals of the fit.
• Figure 5: Detection maps of $\beta$ Pic b with MATISSE. The maps show the reduced $\chi^2$ values after fitting the stellar contamination for each tested planet astrometry $(\Delta\alpha, \Delta\delta)$ in the grid. The grids are centred on MATISSE's pointing during the planet observations. The left map is the size of the pinhole ($1.5\lambda/D$ at 3.5 $\mu$m) and has a resolution of 1 mas. The right map was generated at higher resolution (0.1 mas) on a smaller $20\times20$ mas grid. The most likely astrometry from the MATISSE observations is shown as a blue ellipse, found by fitting a 2D Gaussian curve on the lowest $\chi^2_r$ peak. The planet location and uncertainties predicted by whereistheplanet from the orbital fitting of GRAVITY observations Lacour2021 are shown as a green ellipse. The multiplicity of $\chi^2_r$ peaks is due to the interferometric nature of our data. As coherent flux is periodic as a function of $\vec{\alpha}.\vec{u}$ (see Eq. \ref{['eq:corrFluxRatio']}), an infinity of astrometric solutions can reproduce the signal. This degeneracy gradually disappears when combining data from different baseline lengths and orientations. The $\chi^2_r$ of neighbour peaks can nonetheless still remain close. Using covariances between baselines and wavelengths could help increase the difference between peaks, and will be studied in a future work.