Worlds Next Door: A Candidate Giant Planet Imaged in the Habitable Zone of $\alpha$ Cen A.
I. Observations, Orbital and Physical Properties, and Exozodi Upper Limits

We elected to observe with MIRI and its Four Quadrant Phase Mask Coronagraph (4QPM; Rieke et al., 2015; Wright et al., 2015; Boccaletti et al., 2022) centered at 15.5 $\mu$m (the F1550C filter) for a number of reasons: (1) favorable star-planet contrast ratio for the 200–350 K temperatures expected for a planet heated by $\alpha$ Cen A at 1–3 au; (2) low susceptibility to the effects of wavefront drift at this long wavelength; and (3) the reduced brightness of background objects with typical stellar photospheres. However, despite these advantages, the $\alpha$ Cen AB system presents numerous challenges in planning and executing coronagraphic measurements with JWST at any wavelength.

• •

  The presence of $\alpha$ Cen B only 7″–9″ away from $\alpha$ Cen A puts the full intensity of this bright, [F1550C] $\sim-0.59$ mag star in the focal plane at a position that cannot be attenuated. We developed a strategy ($\S$A.4) to place $\epsilon$ Mus at the position $\alpha$ Cen B would occupy (unocculted) during the observation of $\alpha$ Cen A (occulted). This observation would provide a PSF reference to mitigate the effects of $\alpha$ Cen B.

• •

  The selection of a reference star is complicated by the requirement that it be both comparably bright to $\alpha$ Cen A and have similar photospheric properties in the F1550C waveband.

• •

  The moment-by-moment position of $\alpha$ Cen A is the result of a complex interplay of its high proper motion and parallax (as calculated for the location of JWST at the epoch of observation), and of the orbital motion of $\alpha$ Cen A and $\alpha$ Cen B about their common center of mass (see Akeson et al., 2021, and Table 2).

• •

  With [F1550C] $\sim-1.5$ mag, $\alpha$ Cen A is too bright for direct target acquisition (TA) with MIRI, necessitating a blind offset from a nearby star with an accuracy of $<$10 mas to avoid degradation of the coronagraphic contrast (Boccaletti et al., 2015). The chosen reference star $\epsilon$ Mus ($\S$A.1) is similarly too bright for direct TA, also necessitating a blind offset.

• •

  Offset stars must be of sufficient astrometric accuracy, be as close as possible to $\alpha$ Cen A or $\epsilon$ Mus, but not affected by diffraction or other artifacts from the target stars, and be of sufficient brightness to be readily detectable in a short TA observation at F1000W (Figure 1).

• •

  The time of observation should minimize the change in solar aspect angle between target and reference star observations and thus minimize the change in the telescope’s thermal environment.

• •

  Finally, all MIRI coronagraphic observations using the 4QPM are affected by excess background radiation appearing around the quadrant boundaries referred to as the “Glow Sticks” (Boccaletti et al., 2022).

The above considerations led to the observational sequences described below and detailed in Appendix A. Based on in-flight performance, JWST can place both the target and reference star with an accuracy of $\sim$5–7 mas (1$\sigma$, each axis) behind the MIRI/4QPM. To provide diversity in determining the PSF for post-processing, we selected a 9-point dither pattern for observing the reference star. The multiple reference PSF observations improve the ability of post-processing algorithms to remove residual stellar speckles and help to mitigate wavefront error (WFE) drifts over the 32 hr duration of the entire sequence. The measurement strategy was as follows:

1. 1.

   Offset from a Gaia star ($G9$ in Figure 1) to place $\epsilon$ Mus at the center of the F1550C coronagraphic mask and make a 9-point dithered set of image observations This is followed by observations of a background field to subtract the Glow Stick.

2. 2.

   Place $\epsilon$ Mus at the detector location that $\alpha$ Cen B would occupy in the Roll #1 observation to help mitigate speckles from the unocculted star at the position of $\alpha$ Cen A.

3. 3.

   Offset from a Gaia star ($G0$ or $G5$ in Figure 1) to place $\alpha$ Cen A at the center of the F1550C coronagraphic mask at the Roll #1 V3PA¹¹1V3PA is the position angle (PA) of the V3 reference axis eastward relative to north when projected onto the sky. angle for a sequence of 1250 images, followed by observations of a background field to subtract the Glow Stick.

4. 4.

   Repeat the $\alpha$ Cen A sequence (#3) at a second V3PA angle (Roll #2).

5. 5.

   Repeat the $\epsilon$ Mus 9-point dither sequence (#1) at the mask center.

6. 6.

   Repeat the off-axis $\epsilon$ Mus observation sequence (#2) but at the detector position of $\alpha$ Cen B in the Roll #2 observation.

The observations of $\alpha$ Cen A (Cycle 1 GO, PID #1618; PI: Beichman, Co-PI: Mawet) were initiated in August 2023, but were unsuccessful due to target acquisition and offset failures. A sequence of short test images was obtained in June and July 2024 to validate the target acquisition strategy. Specifically, in July 2024, we executed #1 without dithering and #3 with fewer integrations.²²2No $\epsilon$ Mus reference star observation was acquired at the detector position of $\alpha$ Cen B in the July 2024 test observations. This severely compromised the quality of PSF subtraction. Hence, these observations are not presented. Following successful execution of the test program, we conducted our full-set of science observations in August 2024. We successfully executed steps #1, #4, and #6 (#2 was not part of the sequence planned for this observation). However, the first roll on $\alpha$ Cen A (#3) and the second $\epsilon$ Mus observation (#5) were unsuccessful due to guide star failures.

Based on results from the August 2024 data, the STScI Director’s Office approved a follow-up Director’s Discretionary Time (DDT) program (PID #6797; PI: Beichman, Co-PI: Sanghi). The complete two roll sequence with associated reference star observations (#1–#6) was attempted in February 2025 as part of this DDT program, but due to a telescope pointing anomaly, the first $\alpha$ Cen A roll (#3) was not executed. All other observations were successful.

The STScI Director’s Office approved a second follow-up DDT program (PID #9252; PI: Beichman, Co-PI: Sanghi), which resulted in the successful execution of a full two roll sequence in April 2025. A summary log of all successful observations is provided in Table 8. In all cases, the accuracy of the offsets from Gaia stars was consistent with the expected initial pointing accuracy ($1\sigma$, 5–7 mas), the offset accuracy ($1\sigma$, 1.5 mas) and the line-of-sight jitter ($1\sigma$, 1.5 mas) during the observing sequence at each position. The February 2025 Roll 2 and April 2025 Roll 1 observations showed offsets of $>$ 10 mas from the 4QPM center or from the Eps Mus dither pattern and were thus of lower quality (see dither map in Sanghi & Beichman et al., 2025).

Paper II (Sanghi & Beichman et al., 2025) describes in detail the initial pipeline processing, PSF subtraction techniques for both $\alpha$ Cen A and $\alpha$ Cen B, source identification steps, the photometry and astrometry estimation procedures, and detection sensitivity analysis. Here, we provide a short summary. Level 0 data products were downloaded from MAST, processed for best up-the-ramp calculation of source brightness and bad pixel rejection (Brandt, 2024; Carter et al., 2025), and post-processed to remove the residual stellar diffraction from $\alpha$ Cen A and $\alpha$ Cen B. We assembled distinct reference PSF libraries for each epoch consisting of the individual 400 frames (per dither position) of each 9-pt SGD observation of $\epsilon$ Mus behind the 4QPM and the individual 1250 frames of $\epsilon$ Mus at the unocculted position of $\alpha$ Cen B (for a given roll) obtained at the corresponding epoch. We employed reference star differential imaging (RDI) and jointly subtracted $\alpha$ Cen AB from the 1250 $\alpha$ Cen integrations using the principal component analysis-based Karhunen-Loève Image Processing algorithm (Soummer et al., 2012). Signal-to-noise ratio maps were generated to search for point sources (Mawet et al., 2014) and extended emission (custom method), and assess detection significance.

A comprehensive search of the $\sim$3″ region around $\alpha$ Cen A revealed a single point-like source in the August 2024 data, $S1$ (Figure 2). The source was detected $\approx$1$\farcs$5 East of $\alpha$ Cen A at a S/N between 4–6 (corresponding to a 3.3–4.3$\sigma$ Gaussian significance for the equivalent false positive probability, see Paper II) with a flux density of $\approx 3.5$ mJy (Table 2). The contrast of $S1$ with respect to $\alpha$ Cen A in the F1550C bandpass is $\approx 5.5\times 10^{-5}$. $S1$ is not recovered in the February and April 2025 observations (Figure 2). At wider separations, in all three epochs, we identified two objects denoted KS2 and KS5 that are known from deep 2 $\mu$m VLT/NACO imaging to be background stars (Kervella et al., 2016). In the August 2024 data, KS2 is seen $\approx 6$″ East of $\alpha$ Cen A, after post-processing, exactly in the position expected for a distant, low proper motion star (Figure 2). The bright object KS5 ($K_{s}\sim 7$ mag) is detected just off the edge of the coronagraphic field and will eventually pass within a few mas of $\alpha$ Cen A (mid-2028; Kervella et al., 2016).

Paper II (Sanghi & Beichman et al., 2025) discusses the robustness of the detection of $S1$ and with the help of several tests, presents reasonable evidence that $S1$ is a celestial signal, as opposed to an image artifact. Three primary artifact scenarios are shown to be unlikely:

• •

  $S1$ is not likely a short-lived detector artifact in the $\alpha$ Cen AB integrations. $S1$ was independently detected in multiple subsets of the full 1250 frame integration sequence . Additionally, there was no evidence for transient “hot pixels” in the data, centered on $S1$.

• •

  $S1$ is not likely a PSF-subtraction artifact from the $\epsilon$ Mus coronagraphic reference images. $S1$ was detected in post-processing analyses performed by iteratively excluding each one of the nine dither positions (“leave-one-out” analysis, .

• •

  $S1$ is not likely a PSF-subtraction artifact from imperfect subtraction of $\alpha$ Cen B. $S1$ is well matched to the expected PSF profile and behaves differently with respect to changes in subtraction parameters from another point-like object ($A1$) identified as an artifact from $\alpha$ Cen B.

To assess whether $S1$ is physically associated with $\alpha$ Cen A, we address whether it could be either a background or foreground (Solar System) object. Multiple arguments rule out these scenarios:

• •

  First and most conclusively, the JWST data themselves provide definitive evidence against the hypothesis that $S1$ is a background object. No point source counterparts to $S1$ are detected at the expected location for a background source in the February 2025 and April 2025 observations (Figure 3). See Paper II (Sanghi & Beichman et al., 2025) for further details.

• •

  Archival images taken by Spitzer/IRAC (Rieke & Gautier, 2004), 2MASS, and VLT/NACO (Kervella et al., 2016) when $\alpha$ Cen A was up to one arcminute away from its current position do not show any sources at the $S1$ position. We also considered the effects of interstellar extinction on background source detectability in archival imaging. Extinction maps from Planck and stellar data along the line-of-sight toward $\alpha$ Cen provide a range $20<A_{V}{\rm(mag)}<40$ (Planck Collaboration et al., 2016; Zhang & Kainulainen, 2022), making more extreme $A_{V}$ values unlikely.³³3Planck Extinction maps:
  https://irsa.ipac.caltech.edu/data/Planck/release_2/all-sky-maps/maps/component-maps/foregrounds/COM_CompMap_Dust-DL07-AvMaps_2048_R2.00.fits As shown in Figure 4, if $S1$ were a reddened star (an M0III Kurucz model with $T_{\rm eff}=3800$ K is shown; Buser & Kurucz, 1992) or a normal star-dominated galaxy, its emission would be 4 to 25 times brighter at IRAC wavelengths than at F1550C and would have been detectable by Spitzer, or in the deep NACO $K$-band image. This argument applies to any stellar temperature, since at these wavelengths the emission is approximately Rayleigh-Jeans.

• •

  Figure 4 also shows the spectral energy distribution for a non-photosphere dominated galaxy, the prototypical starburst galaxy or ULIRG, Arp 220, at zero redshift (Polletta et al., 2007). Such an object could have escaped detection in the archival datasets, but the probability of chance alignment with an extragalactic background object is extremely low based on source-counting studies in the MIRI broadband filters. Stone et al. (2024) find that the background density of $F_{\nu}(F1500W)\gtrsim$ 1 mJy sources is $<$0.05 arcmin^-2, corresponding to a chance alignment likelihood $<4\times 10^{-4}$ within a 3″ field-of-view.

  

  

  

• •

  We eliminate the possibility that $S1$ is a foreground Solar System object in a number of ways. An inner main belt asteroid (MBA) at 2.2 AU with a typical temperature of 200 K would have to have a diameter of $>$ 2 km to emit $\sim$ 3 mJy at 15.5 $\mu$m. Such objects are extremely rare, $<10^{-4}$ brighter than 3 mJy at 12 $\mu$m in a 5′$\times$5′ field at $\alpha$ Cen A’s ecliptic latitude of $\beta=-42^{\circ}$ (Brooke, 2003). Furthermore, the completeness for such large MBAs is over 90% and the Minor Planet Catalog shows no known objects at the position of $\alpha$ Cen A at the August epoch⁴⁴4https://minorplanetcenter.net/cgi-bin/checkmp.cgi. Finally, as described in Paper II, there is no angular motion seen between the beginning and the end of $\sim$2.5 hour MIRI observation compared to the expected $>$10 arcsec/hr motion for an MBA at the solar elongation of our observations, $\approx 100^{\circ}$ (Brooke, 2003).

Based on all of the above considerations, we pursue the hypothesis that $S1$ is a planet physically associated with and in orbit around $\alpha$ Cen A, as opposed to an artifact or an astrophysical contaminant, and investigate its properties. Given that $S1$ is only detected in a single roll observation in August 2024, we emphasize that it is, at the moment, a planet candidate. Additional sightings of $S1$ are required with JWST, or other upcoming facilities, to confirm what would be “$\alpha$ Cen Ab”.

We assess our sensitivity to planets around $\alpha$ Cen A across all three epochs of MIRI F1550C observations. Paper II (Sanghi & Beichman et al., 2025) presented the calculation of 2D flux/contrast sensitivity maps. Here, we use the “combined minimum” map from Paper II, which corresponds to the 2D sensitivity map with the best flux sensitivity across all three epochs at each location where the PSF injection-recovery test was performed. We convert the 5$\sigma$ and S/N = 5 flux sensitivities (see Paper II) to effective temperatures ($T_{\rm eff}$) assuming a blackbody model and a typical planet radius of 1 $R_{\rm Jup}$ (for smaller planets, the minimum detectable planet $T_{\rm eff}$ increases). The results are shown in Figure 5. The MIRI F1550C observations are sensitive to $T_{\rm eff}\approx$ 250 K (1 $R_{\rm Jup}$) planets between 1″–2″ for a S/N = 5 detection threshold. Planets colder than 200 K can be detected at wider separations ($>$ 2$\farcs$5). We note here that more realistic planet atmospheric models may have a higher brightness temperature (and thus flux) in the F1550C bandpass relative to the effective blackbody temperature assumed here (see §5.2, for example). This would improve the detectability of colder planets at smaller separations than presented here.

Beichman et al. (2020) predicted that JWST’s ability to resolve the habitable zone around $\alpha$ Cen A would result in unprecedented sensitivity to warm dust—the analog of the thermal emission from dust generated by collisions the asteroid belt in our solar system. We show below that the current observations have not only met but exceeded those expectations with limits as low as a few times the solar system brightness levels at 15.5 $\mu$m.

First, we randomly generate $10^{7}$ orbits matching the astrometry of $S1$ and $C1$ (Table 2) using the Orbits For The Impatient (OFTI) algorithm via the orbitize! package (Blunt et al., 2017, 2020). We apply the default priors in the orbitize! code to the candidate planet’s orbital elements and use a Gaussian prior for $\alpha$ Cen A’s mass and parallax ($M_{\rm A}=1.0788\pm 0.0029\;M_{\odot}$, $\pi=750.81\pm 0.38$ mas, from Akeson et al., 2021). Next, we evaluate the stability of the accepted orbits using the $N$-body simulation software Rebound (Rein and Liu, 2012) over million year timescales using the WHFAST integrator (Rein et al., 2019). A given simulation is deemed unstable for very high planetary eccentricity $(e_{p}>0.95)$ or large planetary distances from the host star $(d_{p}>5\ {\rm au})$. Previous studies showed that orbits that meet either criterion are very likely to be unstable on billion-year timescales (Quarles & Lissauer, 2016, 2018), where more than $90\%$ of orbits stable on a million-year timescale were also stable for billion-year timescales. We find that 30% of the orbits from the initial orbitize! sample are dynamically stable.

We investigate which of the above $S1+C1$ stable orbits are consistent with non-detections in the February and April 2025 observation epochs using the 2D sensitivity maps generated for both epochs in Paper II (Sanghi & Beichman et al., 2025). Specifically, we use the S/N = 5 sensitivity map (rather than the 5$\sigma$ significance sensitivity map) to be stricter in eliminating orbits where $S1+C1$ would have been marginally recovered in our follow-up observations. The sensitivity maps provide the minimum point source flux detectable at a S/N = 5 at different sky coordinates around $\alpha$ Cen A. For each of the stable orbits above, we predict the sky position of the candidate planet in the February and April 2025 observations, and check the corresponding location in the sensitivity map to evaluate whether it would have been detected (for a flux of 3.5 mJy). Orbits where the candidate would have been recovered in either of the two epochs are eliminated. We find that 52% of the stable orbits that fit the $S1+C1$ astrometry are also consistent with non-detections in both February and April 2025 (Figure 8). There is, thus, an a priori significant chance that, if real, the planet candidate could have been missed in both follow-up observation epochs.

The posteriors for the three key orbital elements (semi-major axis $a$, eccentricity $e$, and mutual inclination $i_{\rm mutual}$ with respect to the $\alpha$ Cen AB binary orbit⁵⁵5The mutual inclination is calculated using Equation 19 in Xuan & Wyatt (2020) and using the orbital parameters for $\alpha$ Cen AB in Akeson et al. (2021).) of the stable orbits consistent with the non-detections (Figure 9) show that there are four families of orbits⁶⁶6We do not consider the small fraction ($\sim$0.2% of the total number) of $a<1.5$ au orbits in Figure 9 as they do not agree with $S1$’s relative astrometry within 1$\sigma$ uncertainties.. They correspond to the number of orbital periods that have elapsed between the VLT/NEAR observations in June 2019 and the JWST detection in August 2024 (either $\sim$1.5 or $\sim$2.5 periods, for $a\approx$ 1.6 au and $a\approx$ 2.1 au, respectively) and orbits in either the prograde ($i_{\rm mutual}\approx 50^{\circ}$) or retrograde ($i_{\rm mutual}\approx 130^{\circ}$) direction (Figure 10). In addition to being significantly inclined, the $S1+C1$ planet candidate is in a moderately eccentric ($\approx 0.4$) orbit. A summary of the mean orbital parameters for each family (no RV constraint case) is provided in Table 4. Note that all orbits presented are dynamically stable as evaluated previously (§4.1).

To the family of stable orbits consistent with non-detections, we can add a constraint of $K_{RV}<3$ m s^-1 ($1\sigma$; Butler et al., 2004; Zhao et al., 2018) on the radial velocity of $\alpha$ Cen A, which is also observed in the HARPS RV residuals (Kervella et al., in preparation). This systematic noise floor constrains the minimum mass ($M_{p}\>\mathrm{sin}\>i$) of any planet around $\alpha$ Cen A to be $<100$ $M_{\oplus}$ (2$\sigma$) or $<150$ $M_{\oplus}$ (3$\sigma$) within $\approx 2$ au. Among the dynamically stable $S1+C1$ orbits consistent with the non-detections, assuming $M_{p}=100$ $M_{\oplus}$, we find that 4% of these orbits result in $K_{RV}\leq$ 3 m/s, 50% result in $K_{RV}\leq$ 6 m/s, and 99.8% of all orbits result in $K_{RV}\leq$ 9 m/s (Figure 11, the maximum $K_{RV}$ is $\approx 11$ m/s). Table 4 presents orbital parameters for each case. The semi-major axis and eccentricity remain largely unchanged after applying the RV constraints; however, the mutual and sky inclinations vary as the RV constraint becomes stricter (smaller reflex motion). In summary, the astrometric positions of $S1+C1$ can be fit by dynamically stable orbits consistent with both the non-detections in follow-up observations and existing RV limits. All dynamically stable $S1+C1$ orbits consistent with non-detections can be retrieved from10.5281/zenodo.16280658 (catalog 10.5281/zenodo.16280658).

We use the orbital information to infer the range of plausible effective temperatures for the planet candidate, heated by $\alpha$ Cen A. The equilibrium temperature for a planet on an eccentric orbit depends on the instantaneous stellar input, the planet’s thermal inertia, and radiative timescale. For a gas giant planet, any variations of temperature through the orbit are damped by the thermal inertia of the dense H/He atmosphere (Quirrenbach, 2022). In such a scenario, the correction to the equilibrium temperature to account for changes in the insolation averaged over an eccentric orbit are small, only a few percent, for eccentricities up to 0.5 (Johnson & McClure, 1976; Quirrenbach, 2022). The flux-averaged temperature of a body heated by and orbiting $\alpha$ Cen A in an eccentric orbit, $T_{\rm eq}$, at a distance $d$ is given by

where $T_{\star}=5766$ K (Zhao et al., 2018), $R_{\star}=1.2175\>R_{\odot}$ (Akeson et al., 2021), $A_{B}$ is the Bond albedo, and full heat re-distribution ($f=1$) is assumed. Adopting $A_{B}$ = 0.3 (intermediate between the values for most Hot Jupiters and those of the Solar System gas giants), we compute $T_{\rm eq}$ for all $S1+C1$ stable orbits consistent with the non-detections (§4) and find a bimodal distribution with peaks at $\approx$195 K and $\approx$220 K (Figure 12). The lower temperatures correspond to orbits in the $a>2$ au families and the higher temperatures correspond to orbits in the $a<2$ au families (Table 4).

The contributions of additional sources of heat for the planet candidate’s temperature are negligible compared to stellar insolation. (1) Residual heat of formation: $\alpha$ Cen A is $\sim$5 Gyr old. For a similar internal radiation flux ($F$) as Jupiter or Saturn, $T_{\rm int}<$ 110 K (Li et al, 2018), the increase in temperature is negligible, due to $T\propto F^{1/4}$ and the planet candidate’s higher expected $T_{\rm eq}$. (2) Radiation from $\alpha$ Cen B: $\alpha$ Cen B is less luminous than $\alpha$ Cen A by a factor of $\sim$3 (Akeson et al., 2021) and at the time of JWST observations was $\sim$20 au away from $\alpha$ Cen A. Thus, its contribution to heating is negligible. (3) Tidal heating: the $S1+C1$ candidate is in an eccentric orbit. However, $\dot{E}_{\rm tide}\propto a^{-15/2}$ (Peale & Cassen, 1978) is negligible for $a\sim 2$ au. (4) Heating from radioactivity: this is negligible for Neptune, which is both much colder than $S1+C1$ and likely has a larger fraction of radioactive isotopes.

The goal in this section is to obtain first estimates of the planet candidate’s fundamental parameters ($T_{\rm eff}$, radius, and mass) using atmospheric model grids available in literature for cold planets and brown dwarfs. Given the numerous atmospheric parameter degeneracies involved in fitting two photometric data points, particularly in the observationally unexplored low-mass ($\lesssim$ 200 $M_{\oplus}$), low temperature ($<$ 300 K) planetary regime (see for example, Crotts et al., 2025), we aim only to provide example scenarios that can explain the observed flux measurements. Detailed atmospheric modeling is appropriate for future studies when additional photometry and/or spectroscopy is available (see §6.3).

We jointly fit the F1550C JWST/MIRI flux and the 11.25 $\mu$m VLT/NEAR flux (Table 2), assuming they are related (as indicated by the orbit fits in the previous section). The fitting procedure synthesizes model photometry in the F1550C bandpass ($\approx$15.15–15.85 $\mu$m, using the transmission curve from the SVO filter profile service⁷⁷7http://svo2.cab.inta-csic.es/theory/fps/) and the VLT/NEAR bandpass ($\approx$10–12.5 $\mu$m, constant transmission assumed), and finds the minimum radius ($<$ 1.2 $R_{\rm Jup}$) that yields a model flux consistent with the measured photometry within 1$\sigma$. The effective temperature of the planet is set to 225 K for the atmospheric models fit below, matching that expected for $a<2$ au orbits (Table 4)⁸⁸8We were unable to fit the photometry with 200 K models for planet radii $<$ 1.2 $R_{\rm Jup}$.. We also restrict the surface gravity (log $g$, in CGS units) of the models to 2.5–3.0 dex, chosen to yield a planet mass $\lesssim$150–200 $M_{\rm Earth}$ to be consistent with radial velocity limits ($M_{p}\>\mathrm{sin}\>i<150$ $M_{\oplus}$, 3$\sigma$; inclined orbits can raise the limit on the true planet mass).

ATMO2020++ (Leggett et al., 2021; Meisner et al., 2023): Using the ATMO2020 models with strong vertical mixing as a starting point, ATMO2020++ modifies the adiabatic ideal gas index $\gamma$ (and thus atmospheric temperature gradient) to account for the effect of processes responsible for producing a non-adiabatic cooling curve in giant planet and brown dwarf atmospheres. These processes include complex atmospheric dynamics (e.g., zones, spots, waves) due to rapid rotation, compositional changes due to condensation, upper atmosphere heating by cloud decks or breaking gravity waves, etc. Recent modeling with JWST data has shown that this grid provides an improved fit to Y dwarf spectra compared to the standard-adiabat models (Leggett & Tremblin, 2023, 2024; Luhman et al., 2024). The default ATMO2020++ grid only extends to $T_{\rm eff}=250$ K, so we generated custom models for $T_{\rm eff}=225$ K, log $g=3.0$ dex, and [M/H] $=+0.5,\;+1.0$. We find that the $T_{\rm eff}=225$ K, log $g=3.0$ dex, and [M/H] $=+0.5$ model agrees with the photometry for a radius of 1.1 $R_{\rm Jup}$ (Figure 13). For the above parameters, the planet candidate would have a mass $\approx$150 $M_{\oplus}$.

Sonora and PICASO models: The Sonora Flame Skimmer models (Mang et al., in prep) extend the cloud-free Sonora Elf Owl grid (Mukherjee et al., 2024; Wogan et al., 2025) to colder effective temperatures, lower surface gravities, and a broader range of metallicities. These models incorporate rainout chemistry for H₂O, CH₄, and NH₃—even in cloud-free atmospheres—similar to the treatment in Sonora Bobcat. They also address the underestimation of CO₂ found in the Sonora Elf Owl models (Mukherjee et al., 2024), which has since been revised in Wogan et al. (2025). In addition, we generated a custom grid of cloudy models using PICASO (Batalha et al., 2019; Mukherjee et al., 2023). This grid spans effective temperatures of $T_{\rm eff}=200$ and 225 K, surface gravities of log $g$ = 2.75 and 3.0 dex (cgs), eddy diffusion coefficients $K_{\rm zz}=10^{2}$ and $10^{9}$ cm² s^-1, metallicities of [M/H] = +0.5 and +1.0, and a C/O ratio of 2.5 (relative to solar). Cloudy models have $f_{\rm sed}$ = [4, 6, 8], with H₂O as the only condensing species. We find that the $T_{\rm eff}=225$ K, log $g=2.75$ dex, [M/H] $=+1.0$, log $K_{zz}=9$, C/O = 2.5, $f_{\rm sed}=6$ model agrees with the photometry for a radius of 1.15 $R_{\rm Jup}$ (Figure 13). For the above parameters, the planet would have a mass $\approx$90 $M_{\oplus}$.

Additional models applicable to cool giant planets: We also experimented with fitting the photometry using the Sonora Bobcat cloudless, chemical equilibrium model grid (Marley et al., 2021), the ATMO2020 solar metallicity, disequilibrium chemistry model grid (Phillips et al., 2020), the patchy water cloud models of Morley et al. (2014), and a new grid of self-consistent models by Lacy & Burrows (2023) that incorporate both the effects of water clouds and disequilibrium chemistry. However, we did not find suitable solutions with these grids, as they all required a $T_{\rm eff}\geq 250$ K to fit the photometry for a radius $<1.2$ $R_{\rm Jup}$.

The previous section presented example planet models which can reproduce the brightness of $S1+C1$, but required planet radii $\approx 1.1$ $R_{\rm Jup}$ (driven by the observed F1550C brightness), more commonly observed for hotter planets, but plausible if a rapidly rotating planet is viewed closer to pole-on (the observed surface area can be higher). Alternate explanations for the F1550C brightness include (1) a knot of exozodiacal emission; or (2) a smaller planet with a circumplanetary ring. Given the lack of exozodi detection reported in §3.4, we do not consider the exozodi knot interpretation further, except to note that this would require the knot to dominate the exozodi emission, and for the knot to orbit the star with similar constraints to those reported for the planet scenario (§4) and to have only been detected at one epoch of our observations.

For an interpretation of the emission as circumplanetary material, a straight-forward model is to consider an optically thick ring. A ring is not expected to have significant thermal inertia (as opposed to a gas giant planet, as discussed in §5.1). Thus, the ring temperature at the $S1$ and $C1$ detection epochs will be the instantaneous equilibrium temperature calculated for the true planet-star separation at those epochs. For each stable $S1+C1$ orbit consistent with the non-detections in the prograde $a<2$ au family, we calculate the planet-star separation and the corresponding equilibrium temperature, assuming $A_{B}=0.1$ (similar to asteroids) and $f=1$ (Figure 14). Orbits with $a<2$ au are favored as they yield a higher planet $T_{\rm eq}$, which is required to better fit the F1550C brightness (the prograde and retrograde scenarios yield similar separation distributions and mean values). We find that an optically thick circumplanetary ring would be hotter at the $C1$ epoch ($T_{\rm eq}=257\pm 22$ K) than at the $S1$ epoch ($T_{\rm eq}=209\pm 11$ K).

We modeled the observed $S1+C1$ photometry using various 225 K planet atmospheric models (grids discussed in §5.2) combined with a constant surface area (free parameter) blackbody ring with a temperature of 257 K for the VLT/NEAR 10–12.5 $\mu$m flux and a temperature of $209$ K for the F1550C flux. The photometry agrees, within 1$\sigma$ uncertainties, with a Sonora Flame Skimmer clear, equilibrium, $T_{\rm eff}=225$ K, log $g$ = 3.0 dex, [M/H] = +1.0, and C/O = 1.5 model for a planet radius of 1 $R_{\rm Jup}$ (corresponding to $\approx 120$ $M_{\oplus}$), together with a ring that has a cross-sectional area equivalent to a face-on disk of radius $\approx$ 64,000 km or $\approx$ 0.9 $R_{\rm Jup}$ (Figure 15). This is $\sim$ half the cross-sectional area of Saturn’s rings, which extend to 140,000 km (plus a more tenuous, more distant distribution). Planetary rings lie in their planet’s Roche zone from 1.4–2.5 $R_{p}$, so this explanation seems plausible.

We stress that the ring model discussed above is highly simplified. Geometrical effects make it challenging to develop a fully comprehensive and accurate optically thick ring model. The inclination of the ring with respect to the star affects the ring’s temperature. The inclination of the ring to our line-of-sight together with shadowing of the ring by the planet affects the inferred size and visible emitting area. Additionally, a ring could both shade the planet from starlight, reducing planet temperature, and block planet light towards the observer. In the absence of strong constraints on the planet candidate’s orbit and with only two photometric points, the problem is highly unconstrained. Overall, the key takeaway of the analysis presented above is that a circumplanetary ring around the $S1+C1$ planet candidate is a plausible hypothesis to explain the higher F1550C brightness for a smaller planet than inferred just using atmospheric models. A summary of the inferred mass and radius of the $S1+C1$ candidate from both atmospheric and ring models, as compared with the cold transiting planet population, is presented in Figure 16.

For completeness, we consider an alternative model for circumplanetary material, where the cross-sectional area derived above comes from an optically thin dust distribution, such as one that might arise from the grinding down of a cloud of irregular satellites orbiting the planet (Kennedy et al., 2011). Such a cloud could extend out to roughly half the Hill radius of the planet ($R_{\mathrm{H}}\sim 0.09$ au radius for a low eccentricity, 100 $M_{\oplus}$ planet at 2 au), which would correspond to a distribution with optical depth $\sim 5\times 10^{-5}$. Small grains with realistic optical properties, in models in which the dust is optically thin, are heated above blackbody temperature. This increases the flux in the VLT/NEAR bandpass and makes it more challenging to simultaneously fit the $S1$ and $C1$ photometry, specifically because the ring is expected to have a higher temperature at the $C1$ detection epoch than the $S1$ detection epoch from planet-star separation calculations (Figure 14). This would require the dust distribution to be truncated to avoid the presence of $\mu$m-sized dust. This is in addition to the challenge of retaining the irregular satellites given their expected collisional erosion.

The most pressing task for further work is to capture a second sighting of $S1$ with JWST. Figure 19 identifies an excellent opportunity in August 2026 to recover $S1$ based on the family of stable $S1+C1$ orbits consistent with non-detections described in $\S$4. Around this date, the separation exceeds $\sim$1″ and the predicted location is clear of the 4QPM boundaries. There is an urgency to this given the rapid approach of $\alpha$ Cen A to the known background star denoted KS5 (Kervella et al., 2016). Between mid-2027 and mid-2028, the two will be within 3″ of one another. Additionally, there are numerous opportunities to follow-up the detection of the candidate exoplanet for further characterization with upcoming and future facilities:

• •

  The clear photospheric model has higher flux between 4–5 $\mu$m (Figure 15) compared to the non-adiabatic and cloudy models (Figure 13). NIRCam coronagraphy could serve as a powerful diagnostic tool for the different models presented.

• •

  The coronagraphic instrument on the Nancy Roman Space Telescope has a mask specifically designed to work in the presence of a binary star system (Bendek et al., 2021) and could be used to detect reflected visible light from a gas giant around 1–2 au.

• •

  The METIS instrument on the European Extremely Large Telescope (EELT) should be capable of spectroscopic observations of the $S1$ candidate (Birkby & Parker, 2024) and could even look for radial velocity shifts in the motion of the planet due to the presence of an exomoon.

• •

  Direct mass measurements should be possible with additional RV monitoring and with differential astrometry at millimeter wavelengths with ALMA (Akeson et al., 2021), or at visible wavelengths with the proposed Toliman (Tuthill et al., 2018) or SHERA (J. Christiansen, private comm.) space telescopes.

• •

  Finally, the proposed Habitable Worlds Observatory (HWO) could, if equipped with appropriate binary star rejection capabilities, search for terrestrial-sized planets which might be found within the $\alpha$ Cen A system despite the pessimistic concerns about the stability of orbits exterior to 0.4 au ($\S$6.1.2).