Subject terms: Exoplanets, Atmospheric chemistry

JWST observations and data reduction

We observed Eps Ind A with the JWST/MIRI coronagraphic imager^11,33,34 on 3 July 2023 (JWST General Observer programme 2243; ref.³⁵) in a single sequence. Our observations consisted of science, reference and background images, as detailed in Extended Data Table1. We collected images with two narrowband coronagraphic filters, F1065C and F1550C, which were integrated for 3,772 and 3,922 s, respectively. Each coronagraphic filter has a dedicated four-quadrant phase-mask coronagraph³⁶ to suppress the stellar point spread function (PSF). Dedicated background observations of an empty patch of sky were made at the start of the sequence to allow subtraction of the ‘glow stick’ stray-light feature, first identified in MIRI commissioning³⁴. For each filter and each of the science and reference targets, we collected background observations using the same integration as a single science/reference dither, and we repeated this process over two dither positions. We used the same empty field for all background observations.

We collected PSF reference images of the star DI Tuc (HD211055) for reference differential imaging (RDI)³⁷. We collected five reference images for each filter using the five-point small-grid dither technique^38,39 to account for the imperfect pointing of the JWST (commissioning data suggest a pointing stability of approximately 5–10 mas; ref.⁴⁰). The integration times per dither position were 2,631 and 2,724 s for the F1065C and F1550C observations, respectively. These integrations were chosen to match the peak flux of the Eps Ind A observations at each dither position. In advance of the JWST observations, we also screened the reference star for faint companions to ensure it was suitable for differential imaging. We observed DI Tuc with the Spectro-Polarimetric High-Contrast Exoplanet Research instrument (SPHERE) at VLT on 13 April 2023 (programme 110.25BR, PI E.C.M.) and collected 140 images, each with detector integration time (DIT) = 0.837 s (total exposure time 117 s), in the broadband H filter (λ = 1.6255 µm and Δλ = 0.291 µm). Images were dark- and flat-fielded and bad pixels corrected, and we then aligned all images based on the peak flux position and co-added across the stack of observations. No companions were detected anywhere in the SPHERE field of view (approximately 12″ × 12″). Two companions are in the Gaia catalogue and the JWST field of view but outside the SPHERE field of view (separations 6.5″ and 8.2″). These are sufficiently widely separated that they did not impact the PSF subtraction, though the brighter of these is clearly visible in the reduced data (Extended Data Fig.1).

We used the spaceKLIP pipeline^41,42 for the JWST data analysis. This pipeline provides various high-contrast imaging-specific functionalities for JWST data and, in particular, makes use of the jwst pipeline⁴³ and the pyKLIP Python package⁴⁴ for several key reduction steps. Our reduction follows the steps outlined in detail in ref.⁴² and is briefly summarized here. We started from the stage 0 files (*uncal.fits) generated by the jwst pipeline. We performed ramp-fitting, calibrated the images from detector units to physical units, flat-fielded and subtracted the background images. We then subtracted the stellar PSF using RDI. We explored various parameters. The reductions shown in this work were made using principal component analysis (PCA) with a single subsection and a single annulus. Eight PCA modes were removed.

This process provided excellent starlight suppression for the F1550C images. The contrast performance is provided in Extended Data Fig.2. The stellar PSF is less well matched between the science and reference images for F1065C, probably due to the larger time baseline between these observations (the order of observations is F1065C reference, F1550C reference, F1550C science and F1065C science, as listed in Extended Data Table1). This reduced the planet sensitivity at close separations, and some residual starlight is clearly visible in these images. The full field of view for the 10.65 µm images is shown in Extended Data Fig.1, and the central portion of both images is included in Fig.1. Three astrophysical point sources were detected at F1065C:

1. A bright source to the north-east of Eps Ind A was not previously known. This source was confirmed to be the exoplanet Eps Ind Ab in the current work.

2. A faint source to the west of Eps Ind A is a background star. This source corresponds to Gaia DR3 6412595290591430784 and was also detected in archival Spitzer/IRAC observations.

3. A negative source to the south-east of Eps Ind A is a widely separated companion to DI Tuc. It was subtracted in our RDI reduction. This source corresponds to Gaia DR3 6411654761473726464.

We determined the companion properties using the forward-model PSF fitting routine provided by pyKLIP and implemented in spaceKLIP. We generated a PSF template at the location of the companion using webbpsf_ext. We then fitted the location and intensity of this PSF to the data to measure the photometry and astrometry of the observations; these values are listed in Extended Data Table2.

We also reduced the data using non-negative matrix factorization^45,46 to verify our results. Non-negative matrix factorization involves decomposing an image matrix into two non-negative matrices, one has features and the other weights for reconstructing the original matrix. The features matrix extracted from the RDI PSF library was used to model the stellar PSF in the target image frames. PSF subtraction was then performed to image the exoplanet. We performed the non-negative matrix factorization reduction using all frames from the DI Tuc PSF library to construct components. For F1065C, this involved using 605 frames to construct 605 components. For F1550C, the number of frames and components was 65. We tried alternative approaches, including using the entire PSF library with fewer components and using a smaller PSF library with components selected using their closeness to the target frames (using the Euclidean norm), but using the entire library with the maximum number of components consistently produced the best images.

Background constraints derived from archival data products

Our analysis of archival data provided tight constraints on a possible background contaminant. If the point source were an extragalactic contaminant, it would have no proper motion. It would be found 4.1″ to the north-east of the JWST Eps Ind A position, namely at right ascension 22 h 03 min 33.17 s and declination −56° 48′ 06.0″. For a background object, these absolute right ascension and declination coordinates would not change as Eps Ind A moves across the sky at a rate of 4.7″ yr⁻¹ and would be well resolved from the star in sufficiently old archival data (allowing for a sufficient movement of the foreground star, relative to the background position). Even if a background contaminant were a distant object within the Milky Way, it would have a proper motion of at most a few milliarcseconds per year and at most a few tens of milliarcseconds over the approximately 2 decades spanned by the data products presented here. We searched the Gaia¹⁰ and 2MASS⁴⁷ catalogues and reanalysed data from the IRAC and the multiband imaging photometer (MIPS) onboard Spitzer, but we did not identify any background sources consistent with the point source.

Spitzer observations provide excellent background constraints and bracket the wavelengths of our JWST observations. Several Spitzer observations of Eps Ind A were carried out during 2004. Eps Ind A had moved approximately 90'' in the 19 yr between the Spitzer and JWST observations. Eps Ind A was observed with Spitzer/IRAC on 1 May 2004 (programme ID 90; PI M. Werner), with data collected in all four IRAC channels (nominally 3.6, 4.5, 5.8 and 8.0 µm). The star was used primarily as a PSF reference star for a programme studying the Eps Eri debris disk^15,48,49. We reanalysed these data, starting from the science-ready mosaic files from the Spitzer Heritage Archive for each of the four wavelength channels. Even though the background position was well separated from the star at this epoch, there was still a significant signal in the PSF wings (Extended Data Fig.3). We used the Eps Eri observations from the same programme as a PSF reference for RDI and for subtracting the stellar PSF wings, thereby allowing the best possible contrast at the background location. We used a grid-based approach to optimize the spatial (x,y) location and flux scaling of the Eps Eri images and, thereby, to provide the best match to the PSF wing structure of the Eps Ind images. We then subtracted this best-matching image from the Eps Ind data to produce the RDI-subtracted images. The science-ready images and RDI-subtracted images are shown in Extended Data Fig.3. The position of Eps Ind A during the Spitzer and JWST epochs is highlighted. We calculated the flux limits based on the pixel variance in the RDI-subtracted image, with known sources masked, and validated our limits by performing aperture photometry on identified sources in the image.

Eps Ind A was also observed with Spitzer/MIPS on 13 October 2004 (programme ID 41; PI G. Rieke)⁵⁰. Data were collected in all three MIPS channels (nominally 24, 70 and 160 µm). We analysed the science-ready mosaic files from the Spitzer Heritage Archive. Eps Ind A is unsaturated in the MIPS images, and the JWST position is well resolved from the star (Extended Data Fig.4) and within the field of view for the 24 and 70 µm channels. We calculated the flux upper limit based on the standard deviation of pixel values in a large box around the position of Eps Ind A, excluding pixels within 1 arcmin of the star. No sources were observed within 10″ of the JWST Eps Ind A position in the Spitzer IRAC 8.0 µm images or in any of the MIPS images. One source was identified approximately 9″ west of Eps Ind A in the 3.6, 4.5 and 5.8 µm images. This source was also identified in the JWST/MIRI images and is labelled 2 in Extended Data Fig.1. The location and magnitude measurements correspond to that of the star Gaia DR3 6412595290591430784. This is a background object and is not associated with Eps Ind A.

A background spectrum for the point source is shown in Extended Data Fig.5. This indicates the flux and upper limits that would apply to the object if it were a stationary background object or a background object that had moved by less than a few hundred milliarcseconds per year. That is, these upper limits do not apply to the planet, which has the same proper motion as Eps Ind A and was not present at this location in the background observations. The 8.0 µm images provide a particularly stringent constraint on the background scenario. We ruled out background objects to 16.07 mag at this wavelength, whereas the point source is 13.16 mag at 10.65 µm. A background object would have to be extremely red to be compatible with both measurements.

Reanalysis of archival VISIR/NEAR data

We revisited archival VISIR/NEAR^51,52 observations of Eps Ind A. We started from previously published reduced datacubes¹⁶, for which frames were calibrated and aligned. We removed low-quality frames, and then binned and spatially high-pass filtered the remaining frames. The NEAR sensitivity was optimized for the small separation (0.5″–3.0″) around the stars. Ground-based thermal infrared observations require fast chopping to quickly subtract the varying sky background and excess low-frequency detector noise. With NEAR, chopping was done using the deformable secondary mirror of the VLT, which provides a maximum chop throw of about 4.5″. The reduction with angular differential imaging (ADI) relies on subtracting images taken at a different field orientation, which leaves residuals of real sources in the azimuthal direction. As the separation of the imaged planet from the host star was close to the chop throw, these residuals masked the planet if ADI were used for data reduction. To avoid this effect, we simply co-added data without performing ADI, as the expected separation was well within the background-limited regime rather than the contrast-limited regime. We also performed a locally optimized combination of images (LOCI)⁵³ PCA ADI analysis with three principal components in a small patch around the expected companion location. This helped to improve the planet signal and suppress the thermal background. In each case, we also convolved the final images with a top hat to enhance any true signals in the data. Both images are shown in Fig.2. To confirm the source, we also looked at the companion position relative to the off-axis PSFs. We centred, aligned and derotated the frames relative to this position. We found that the planet signal was also present in the final averaged convolved frames relative to the off-axis PSFs, negating the possibility of the source being a speckle or a ghost from internal reflections. We assumed an error of half the full-width at half-maximum on the companion position, due to the high-spatial resolution and sampling of the NEAR. The planet was at a separation of 4.82 ± 0.16″ and at a position angle of 40–45° anticlockwise of north. To estimate the flux, we used planet injection and recovery tests. As the planet signal was limited by the thermal background, we put a conservative range of (4–8) × 10⁻⁵ on contrast and a range of 0.18–0.35 mJy on flux. These values are also included in Extended Data Table2.

Orbital constraints

Our goal in targeting Eps Ind A with JWST coronagraphic imaging was to detect a known, massive planet for which dynamical mass information was accessible. However, the imaged planet is inconsistent with the previously claimed planet in the system. The planet has a significantly larger mass and semimajor axis than the claimed Eps Ind Ab (refs.^2,8,9) and a strikingly different on-sky position than predicted. It is clear that the imaged planet imparts a significant, long-term, radial-velocity acceleration on the host star, although it was initially unclear whether this is the only planet in the system.

We attempted to fit the orbit of the imaged planet by assuming that it is, indeed, the only planet in the system and that it is responsible for all the dynamical measurements (radial-velocity data and astrometry) of the host star. We used radial-velocity data from the long camera (LC) and very long camera (VLC) of the coudé echelle spectrometer at the European Southern Observatory (ESO)⁵, from the ultraviolet and visual echelle spectrograph (UVES)² and from the high accuracy radial-velocity planet searcher (HARPS), using the data reduction presented in ref.⁵⁴. Eps Ind A has been targeted intensively with HARPS, and we binned any data points collected within a single night with HARPS. We treated the HARPS data as three separate instruments, to account for any possible baseline shifts during an instrument upgrade in 2015 and a telescope shutdown in 2020. This gave us 493 data points, from six independent instruments, covering from 3 November 1992 to 29 December 2021. We fitted orbits using orvara⁵⁵, considering the 493 radial-velocity points, the two direct imaging epochs and the Hipparcos–Gaia astrometry of the host star, and using a host star mass prior of 0.76 ± 0.04 M_⊙. This process converged to an orbit with mass${6.31}_{-0.56}^{+0.60}$ M_J, semimajor axis${28.4}_{-7.2}^{+10}\mathrm{au}$, eccentricity${0.40}_{-0.18}^{+0.15}$ and inclination 103.7 ± 2.3° (Extended Data Fig.6). Without the NEAR astrometry, we found an orbit with mass${6.17}_{-0.57}^{+0.65}$ M_J, semimajor axis${31}_{-10}^{+13}\mathrm{au}$, eccentricity${0.47}_{-0.23}^{+0.15}$ and inclination 104.0 ± 2.4°. These orbits are highly consistent with each other, further validating that the NEAR epoch is a positive redetection of the companion. These orbits are also consistent with all in-hand dynamical data. Curiously, several previous works had derived properties of the claimed planet Eps Ind Ab and found consistent results^2,8,9, but these results are inconsistent with the planet observed in this work. This may be due to overfitting of the in-hand data, as fitting accurate orbits with insufficient orbital phase coverage is notoriously hard⁵⁶, or it may hint at another component in the system that biased the previous one-planet fits. The consistency between the fits with and without the NEAR astronomy points towards a one-planet system. We leave a detailed exploration of the discrepancy between the imaged planet and the previous radial-velocity solutions to a future work, but note that the dynamical mass derived in this work is valid only if Eps Ind Ab is the only massive planet present.

Stellar photometry

We used BT-NextGen models^57,58 to predict the photometry of the host star Eps Ind A in the JWST/MIRI and VLT/NaCo filters. We fitted spectra to the in-hand photometry from TYCHO, Gaia, 2MASS, WISE, Spitzer/MIPS and the photoconductor array camera and spectrometer (PACS) onboard Herschel^{10,47,50,59,60}. We allowed the effective temperature, log(g), metallicity and radius to vary and held the stellar distance fixed at 3.639 pc. We sampled the posterior with a Monte Carlo Markov chain (using emcee; ref.⁶¹) to derive the best-fitting spectrum and uncertainty. The best-fitting values are effective temperatureT_eff = 4,760 ± 15 K, log(g) = ${5.25}_{-0.34}^{+0.18}$, [Fe/H] = 0.22 ± 0.12 andR = 0.679 ± 0.004 R_⊙, although we caution that these values are driven by photometry only, and spectroscopic fits of the stellar parameters should be used over these values when interpreting the metallicity and log(g) of the host star. We then derived model photometry and uncertainties by selecting samples from the Markov chain Monte Carlo (MCMC) posterior and integrating over the filter profiles. The best-fitting spectrum as well as the measured (green and pink) and model (blue) photometry of Eps Ind A is shown in Extended Data Fig.7. We used this model to derive all companion photometry and flux limits in this work, including reconverting archival contrast curves to flux upper limits with this model.

The WISE observations of Eps Ind A appear to be unreliable. Eps Ind A is significantly saturated in the WISE W1, W2 and W3 observations (the WISE source catalogue⁶⁰ lists saturation fractions of w1sat = 0.253, w2sat = 0.239 and w3sat = 0.144). For this reason, we did not include the W1, W2 and W3 magnitudes directly in our fitting process. Our derived stellar magnitudes show some discrepancy with those used to derive previous L′ limits¹⁷, which used the WISE W1 magnitude of 2.9 as a proxy for the L′ magnitude of Eps Ind A. Our stellar fit indicates a somewhat brighter magnitude for the star of 2.119 ± 0.008 and 2.115 ± 0.009 for the W1 and L′ filters respectively. The only published L-band magnitude for Eps Ind isL = 2.12 (ref.⁶²), consistent with our derived W1 and L′ magnitudes.

Extended data figures and tables

Peer review information

Nature thanks Ignas Snellen and the other, anonymous, reviewer(s) for their contribution to the peer review of this work.

• 1.NASA Exoplanet Archive.IPAChttps://exoplanetarchive.ipac.caltech.edu/ (accessed 22 June 2024).
• 2.Feng, F. et al. Detection of the nearest Jupiter analogue in radial velocity and astrometry data.Mon. Not. R. Astron. Soc.490, 5002–5016 (2019). [Google Scholar]
• 3.Chen, M. et al. Precise dynamical masses of Eps Indi Ba and Bb: evidence of slowed cooling at the L/T transition.Astron. J.163, 288 (2022). [Google Scholar]
• 4.Endl, M. et al. The planet search program at the ESO coudé echelle spectrometer. III. The complete long camera survey results.Astron. Astrophys.392, 671–690 (2002). [Google Scholar]
• 5.Zechmeister, M. et al. The planet search programme at the ESO CES and HARPS. IV. The search for Jupiter analogues around solar-like stars.Astron. Astrophys.552, A78 (2013). [Google Scholar]
• 6.Brandt, T. D. The Hipparcos–Gaia catalog of accelerations.Astrophys. J. Suppl. Ser.239, 31 (2018).
• 7.Kervella, P., Arenou, F., Mignard, F. & Thévenin, F. Stellar and substellar companions of nearby stars from Gaia DR2. Binarity from proper motion anomaly.Astron. Astrophys.623, A72 (2019). [Google Scholar]
• 8.Feng, F., Butler, R. P., Vogt, S. S., Holden, B. & Rui, Y. Revised orbits of the two nearest Jupiters.Mon. Not. R. Astron. Soc.525, 607–619 (2023). [Google Scholar]
• 9.Philipot, F., Lagrange, A. M., Rubini, P., Kiefer, F. & Chomez, A. Updated characterization of long-period single companion by combining radial velocity, relative astrometry, and absolute astrometry.Astron. Astrophys.670, A65 (2023). [Google Scholar]
• 10.Gaia Collaborationet al. Gaia Data Release 3. Summary of the content and survey properties.Astron. Astrophys.674, A1 (2023). [Google Scholar]
• 11.Rieke, G. H. et al. The mid-infrared instrument for the James Webb Space Telescope. I. Introduction.Publ. Astron. Soc. Pac.127, 584 (2015). [Google Scholar]
• 12.Wu, C. K. W. et al. Source counts at 7.7–21 µm in CEERS field with JWST.Mon. Not. R. Astron. Soc.523, 5187–5197 (2023). [Google Scholar]
• 13.Fulton, B. J. et al. California Legacy Survey. II. Occurrence of giant planets beyond the ice line.Astrophys. J. Suppl. Ser.255, 14 (2021).
• 14.Vigan, A. et al. The SPHERE infrared survey for exoplanets (SHINE). III. The demographics of young giant exoplanets below 300 au with SPHERE.Astron. Astrophys.651, A72 (2021). [Google Scholar]
• 15.Marengo, M. et al. A Spitzer IRAC search for substellar companions of the debris disk star Eps Eridani.Astrophys. J.647, 1437–1451 (2006). [Google Scholar]
• 16.Pathak, P. et al. High-contrast imaging at ten microns: a search for exoplanets around Eps Indi A, Eps Eri, Tau Ceti, Sirius A, and Sirius B.Astron. Astrophys.652, A121 (2021). [Google Scholar]
• 17.Viswanath, G. et al. Constraints on the nearby exoplanet Eps Indi Ab from deep near- and mid-infrared imaging limits.Astron. Astrophys.651, A89 (2021). [Google Scholar]
• 18.Janson, M. et al. Imaging search for the unseen companion to Eps Ind A – improving the detection limits with 4 µm observations.Mon. Not. R. Astron. Soc.399, 377–384 (2009). [Google Scholar]
• 19.Burrows, A. et al. A nongray theory of extrasolar giant planets and brown dwarfs.Astrophys. J.491, 856–875 (1997). [Google Scholar]
• 20.Cushing, M. C. et al. The discovery of Y dwarfs using data from the Wide-field Infrared Survey Explorer (WISE).Astrophys. J.743, 50 (2011). [Google Scholar]
• 21.Luhman, K. L. et al. JWST/NIRSpec observations of the coldest known brown dwarf.Astron. J167, 5 (2024). [Google Scholar]
• 22.Morley, C. V. et al. Water clouds in Y dwarfs and exoplanets.Astrophys. J.787, 78 (2014). [Google Scholar]
• 23.Lacy, B. & Burrows, A. Self-consistent models of Y dwarf atmospheres with water clouds and disequilibrium chemistry.Astrophys. J.950, 8 (2023). [Google Scholar]
• 24.Mukherjee, S. et al. The Sonora substellar atmosphere models. IV. Elf Owl: atmospheric mixing and chemical disequilibrium with varying metallicity and C/O ratios.Astrophys. J.963, 73 (2024). [Google Scholar]
• 25.Fegley, J. & Lodders, K. Atmospheric chemistry of the brown dwarf Gliese 229B: thermochemical equilibrium predictions.Astrophys. J. Lett.472, L37 (1996). [Google Scholar]
• 26.Öberg, K. I., Murray-Clay, R. & Bergin, E. A. The effects of snowlines on C/O in planetary atmospheres.Astrophys. J. Lett.743, L16 (2011). [Google Scholar]
• 27.Phillips, M. W. et al. A new set of atmosphere and evolution models for cool T-Y brown dwarfs and giant exoplanets.Astron. Astrophys.637, A38 (2020). [Google Scholar]
• 28.Scholz, R. D., McCaughrean, M. J., Lodieu, N. & Kuhlbrodt, B. ε Indi B: a new benchmark T dwarf.Astron. Astrophys.398, L29–L33 (2003). [Google Scholar]
• 29.McCaughrean, M. J. et al. Eps Indi Ba,Bb: the nearest binary brown dwarf.Astron. Astrophys.413, 1029–1036 (2004). [Google Scholar]
• 30.Mullally, S. E. et al. JWST directly images giant planet candidates around two metal-polluted white dwarf stars.Astrophys. J. Lett.962, L32 (2024). [Google Scholar]
• 31.Bardalez Gagliuffi, D. C. et al. Solving a solar neighborhood crime scene by imaging 14 Her c.JWST ProposalCycle 2, 3337 (2023).
• 32.Matthews, E. C. et al. Super-Jupiters in our backyard: MIRI coronagraphic imaging of a massive planet/brown dwarf orbiting an M dwarf at 12pc.JWST ProposalCycle 3, 5229 (2024).
• 33.Boccaletti, A. et al. The mid-infrared instrument for the James Webb Space Telescope. V. Predicted performance of the MIRI coronagraphs.Publ. Astron. Soc. Pac.127, 633 (2015). [Google Scholar]
• 34.Boccaletti, A. et al. JWST/MIRI coronagraphic performances as measured on-sky.Astron. Astrophys.667, A165 (2022). [Google Scholar]
• 35.Matthews, E. C. et al. A direct detection of the closest Jupiter analog with JWST/MIRI.JWST ProposalCycle 1, 2243 (2021).
• 36.Rouan, D., Riaud, P., Boccaletti, A., Clénet, Y. & Labeyrie, A. The four-quadrant phase-mask coronagraph. I. Principle.Publ. Astron. Soc. Pac.112, 1479–1486 (2000). [Google Scholar]
• 37.Smith, B. A. & Terrile, R. J. A circumstellar disk around β Pictoris.Science226, 1421–1424 (1984). [DOI] [PubMed] [Google Scholar]
• 38.Soummer, R., Hines, D. & Perrin, M.Simulations of Target Acquisition with MIRI Four-quadrant Phase Mask Coronagraph (I), Report JWST-STScI-003063, SM-12 (STSI, 2013).
• 39.Lajoie, C.-P. et al. Small-grid dithers for the JWST coronagraphs.Proc. SPIE9904, 99045K (2016).
• 40.Rigby, J. et al. The science performance of JWST as characterized in commissioning.Publ. Astron. Soc. Pac.135, 048001 (2023). [Google Scholar]
• 41.Kammerer, J. et al. Performance of near-infrared high-contrast imaging methods with JWST from commissioning.Proc. SPIE12180, 121803N (2022).
• 42.Carter, A. L. et al. The JWST early release science program for direct observations of exoplanetary systems. I. High-contrast imaging of the exoplanet HIP 65426 b from 2 to 16 µm.Astrophys. J. Lett.951, L20 (2023). [Google Scholar]
• 43.Bushouse, H. et al. JWST calibration pipeline. Zenodo 10.5281/zenodo.6984365 (2024).
• 44.Wang, J. J. et al. pyKLIP: PSF subtraction for exoplanets and disks. Astrophysics Source Code Library, record ascl:1506.001 (ASCL, 2015).
• 45.Ren, B., Pueyo, L., Zhu, G. B., Debes, J. & Duchêne, G. Non-negative matrix factorization: robust extraction of extended structures.Astrophys. J.852, 104 (2018). [Google Scholar]
• 46.Krishanth P M, S. et al. NMF-based GPU accelerated coronagraphy pipeline.Proc. SPIE12680, 1268021 (2023).
• 47.Skrutskie, M. F. et al. The Two Micron All Sky Survey (2MASS).Astron. J.131, 1163–1183 (2006). [Google Scholar]
• 48.Marengo, M. et al. Spitzer/Infrared array camera limits to planetary companions of Fomalhaut and Epsilon Eridani.Astrophys. J.700, 1647–1657 (2009). [Google Scholar]
• 49.Backman, D. et al. Epsilon Eridani’s planetary debris disk: structure and dynamics based on Spitzer and Caltech submillimeter observatory observations.Astrophys. J.690, 1522–1538 (2009). [Google Scholar]
• 50.Gáspár, A., Rieke, G. H. & Balog, Z. The collisional evolution of debris disks.Astrophys. J.768, 25 (2013). [Google Scholar]
• 51.Lagage, P. O. et al. Successful commissioning of VISIR: the mid-infrared VLT instrument.Messenger117, 12–16 (2004). [Google Scholar]
• 52.Kasper, M. et al. NEAR: low-mass planets in α Cen with VISIR.Messenger169, 16–20 (2017). [Google Scholar]
• 53.Lafrenière, D, Marois, C., Doyon, R., Nadeau, D. & Artigau, É. A new algorithm for point-spread function subtraction in high-contrast imaging: a demonstration with angular differential imaging.Astrophys. J.660, 770 (2007).
• 54.Trifonov, T. et al. Public HARPS radial velocity database corrected for systematic errors.Astron. Astrophys.636, A74 (2020). [Google Scholar]
• 55.Brandt, T. D. et al. orvara: an efficient code to fit orbits using radial velocity, absolute, and/or relative astrometry.Astron. J.162, 186 (2021). [Google Scholar]
• 56.Lagrange, A. M. et al. Radial distribution of giant exoplanets at Solar System scales.Astron. Astrophys.677, A71 (2023). [Google Scholar]
• 57.Allard, F., Homeier, D. & Freytag, B. Model atmospheres from very low mass stars to brown dwarfs. InProc. 16th Cambridge Workshop on Cool Stars, Stellar Systems, and the Sun (eds Johns-Krull, C., Browning, M. K. & West, A. A.)448, 91 (2011).
• 58.Allard, F., Homeier, D. & Freytag, B. Models of very-low-mass stars, brown dwarfs and exoplanets.Philos. Trans. R. Soc. Lond. Ser. A370, 2765–2777 (2012). [DOI] [PMC free article] [PubMed] [Google Scholar]
• 59.Høg, E. et al. The Tycho-2 catalogue of the 2.5 million brightest stars.Astron. Astrophys.355, L27–L30 (2000). [Google Scholar]
• 60.Cutri, R. M. et al.VizieR Online Data Catalog: WISE Preliminary Data Release (Cutri+ 2011) (ADS, 2012).
• 61.Foreman-Mackey, D., Hogg, D. W., Lang, D. & Goodman, J. emcee: the MCMC hammer.Publ. Astron. Soc. Pac.125, 306 (2013). [Google Scholar]
• 62.Morel, M. & Magnenat, P. UBVRIJKLMNH photoelectric photometric catalogue. (Magnetic tape).Astron. Astrophys. Suppl. Ser.34, 477 (1978). [Google Scholar]

Data Availability Statement

All observational data in this work are available through public data archives. In particular, the JWST data were collected through the General Observer programme 2243 (PI E.C.M.) and are available through the Mikulski Archive for Space Telescopes (https://mast.stsci.edu). Spitzer data were collected through programmes 90 (PI M. Werner) and 41 (PI G. Rieke) and are available through the Spitzer Heritage Archive (https://irsa.ipac.caltech.edu/applications/Spitzer/SHA/). VLT data from VISIR/NEAR and SPHERE were collected through programmes 60.A-9107/103.201H (PI M. Meyer) and 110.25BR (PI E.C.M.), respectively, and are available through the ESO science archive (http://archive.eso.org/cms.html). Radial-velocity data and atmospheric models were taken from previously published literature, as indicated in the text.

Data reduction was carried out using the publicly available spaceKLIP package (https://github.com/kammerje/spaceKLIP), which also uses functions from the jwst pipeline (https://jwst-pipeline.readthedocs.io), the pyKLIP Python package (https://pyklip.readthedocs.io) and the webbpsf_ext package (https://webbpsf.readthedocs.io;https://github.com/JarronL/webbpsf_ext). Orbit fitting was carried out using the publicly available orvara package (https://github.com/t-brandt/orvara). We used various functions from astropy, pandas, matplotlib and seaborn in the analysis and to create figures.