Imaging low-mass planets within the habitable zone of α Centauri

The study targets direct imaging of potentially Earth-like, temperate exoplanets around nearby stars, focusing on α Centauri, the nearest stellar system whose Sun-like components have habitable zones near ~1 au (~1 arcsecond at 1.3 pc). While direct imaging to date has revealed young, self-luminous super-Jovian planets at large separations and short wavelengths (≤5 µm), imaging mature, colder planets in habitable zones requires mid-infrared observations (~10–20 µm) where such planets are brightest. Challenges include high thermal background and coarser diffraction-limited resolution. Radial-velocity surveys exclude only relatively high-mass planets in the α Centauri A and B habitable zones (≥53 M⊕ for A; ≥8.4 M⊕ for B), leaving room for lower-mass, dynamically stable planets. The NEAR experiment aims to demonstrate that low-mass, habitable-zone exoplanets can be imaged from the ground in a practical observing program (~100 h), assessing sensitivity and potential detections around α Centauri.

Prior direct imaging has focused on young, massive exoplanets at wide separations with instruments operating at ≤5 µm, where thermal background is lower but temperate planets are faint. Landmark detections include multi-planet systems like HR 8799 and β Pictoris b, enabling atmospheric and orbital studies. Statistical surveys (e.g., LEECH, GPI, SPHERE/SHINE) constrained occurrence rates of giant planets at >10 au. For nearby stars, α Centauri presents uniquely favorable angular scales for habitable-zone imaging. RV studies constrain massive planets in α Centauri but allow lower-mass planets to persist; Proxima Centauri hosts at least two >Earth-mass planets detected via RV. Proposed future facilities and concepts (ELT/METIS, LUVOIR, HabEx) forecast capabilities to image and characterize habitable-zone planets. Techniques like angular differential imaging (ADI), PCA/KLIP, and coronagraphy (e.g., AGPM vortex) underpin high-contrast imaging, while mid-IR detectors face challenges like excess low-frequency noise (ELFN). The NEAR program builds upon these advances by upgrading VISIR with mid-IR coronagraphy, shaped pupils, and adaptive optics via the VLT’s deformable secondary mirror to mitigate thermal background and enable mid-IR high-contrast performance.

Observational campaign and instrument: The NEAR experiment upgraded the VLT mid-IR instrument VISIR to enable high-contrast imaging at 10–12.5 µm (N band). Upgrades included an AGPM vector vortex coronagraph optimized for mid-IR, a shaped-pupil mask to control off-axis starlight, and use of the VLT Unit Telescope 4’s deformable secondary mirror (DSM) to provide adaptive optics (AO) without adding warm optics. The DSM also performed chopping between α Cen A and B at ~8–10 Hz; subtracting alternating star positions reduced the thermal background and partially mitigated AGPM glow and residual coronagraphic PSF. Observations were conducted in pupil-stabilized mode.

Data acquisition: α Centauri was observed 2019-05-23 to 2019-06-11, with an additional night on 2019-06-27 (not combined due to orbital motion concerns but used for astrometric checks). Total exposure time ~100 h; 23 h discarded due to high background, coronagraph misalignment, or AO issues. The final dataset used 76.9 h of good-quality data. VISIR spectral filter transmitted 10–12.5 µm; FWHM ~0.28″ (~6 pixels). Detector integration time (DIT) was 6 ms (5.5 ms on 2019-05-24, normalized in processing); eight frames were averaged and two skipped during chopping transitions, yielding an 8.33 Hz chopping frequency.

Primary data reduction pipeline: For frames with α Cen A on the coronagraph, the mean of two neighboring frames (with α Cen B) was subtracted (chop subtraction) to remove ELFN and background structure (including AGPM glow) and to attenuate the residual coronagraphic PSF. Five-hundred-frame cubes were coadded into 24 s images and assembled into nightly data cubes. Frames were aligned via the unocculted PSF of α Cen B and the A-B coronagraphic residual center determined by rotational centering. Quality control rejected ~10% of frames whose maximum cross-correlation to a running mean (computed over 5–45 pixel radii) was <0.9. Known detector artifacts (detector persistence stripes from chopping, and optical ghost arcs from the dichroic/filter) were modeled and subtracted. Destriping subtracted the mode of each row and column. High-pass filtering subtracted a 15-pixel running-median–smoothed version of each frame. Original frames were stacked into 360 s images, then processed with classical ADI and KLIP (four KL modes in an annulus 5–45 pixels). A second high-pass filter (same parameters) reduced residual low-frequency structure. Nightly images were combined pixelwise using noise-weighted ADI, and nightly results were variance-weighted averaged into the final mosaics.

Secondary pipeline: Similar preprocessing but without artifact modeling/subtraction (used for cross-checks). Quality metrics per chopped image included AO performance, coronagraphic leakage, and sky-background variance; 79.3 h remained after quality cuts. Frames were co-aligned to the midpoint between off-axis α Cen A and B positions and mean-combined into 60 s frames. An ADI-based PCA model was computed over an annulus (optimized with fake planet injections) using 15 principal components, inner radius 8 px, outer radius 16 px (with further regions also processed). Results were variance-weighted combined per night. Both pipelines provided comparable performance before artifact subtraction; the primary pipeline’s artifact modeling yielded higher fidelity final images.

Sensitivity assessments: Background-limited sensitivity was estimated from the standard deviation of pixel intensities within a 0.35″ (~8 px) aperture far from the stars, yielding ~1.67×10⁻⁷ contrast relative to α Cen A (~22 µJy). Pixel noise increases near the AGPM glow (approximately doubled standard deviation at 1″ separation). Empirical detection limits were established by forward-modeling injections of simulated point sources with planetary brightnesses throughout the processing chain; identifiable detections required signals ~an order of magnitude above the background-noise curve, corresponding to ~2–3×10⁻⁶ contrast at ~1″. Completeness was computed via Monte Carlo sampling of orbital parameters and radii, assuming Bond albedo Ag=0.3 and internal heating contributing 10% or 50% of equilibrium temperature, with inclination prior P(i)∝sin i.

Instrumental/observing specifics: AO Strehl ratios typically >97%. Chopping half-cycle 60 ms; 8.33 Hz. The Lyot stop was chromium on the NEAR spectral filter. One beam diameter corresponds to ~14° azimuth at 1″; sky rotation during 360 s produced only ~2.2° smearing, well below a beam. Data were high-pass filtered twice and destriped to suppress low-frequency and row/column fixed-pattern noise. Artifact modeling addressed detector persistence from both stars and negative arcs from optical ghosts. Two independent pipelines and fake-planet tests validated results. All data are publicly available (ESO archive program 2102.C-5011(A)).

• Sensitivity: Background-limited pixel-noise floor corresponds to ~1.67×10⁻⁷ contrast relative to α Cen A (~22 µJy) in empty regions; near 1″ separation, AGPM glow approximately doubles pixel-to-pixel noise.
• Empirical detection limit: Simulated planet injections indicate identifiable point sources at ~2–3×10⁻⁶ contrast relative to α Cen A at 1″ (SNR3 in 76.9 h). This enables sensitivity to warm sub-Neptune/Saturn-sized planets across much of α Cen A’s habitable zone and to Jupiter-sized planets (with modest additional heat or low albedo) around α Cen B.
• One-hour performance: In background-limited regions, sensitivity ~0.75 mJy in 1 h.
• Final contrast-limited sensitivity near 1″ (SNR=3, 77 h): ~3×10⁻⁶ contrast to α Cen A (~0.4 mJy).
• Completeness: Maximum ~80% for Jovian-radius planets, ~85% for slightly larger (inflated) planets; ~1–10% chance to detect a warm ~3 R⊕ planet around α Cen A at the extreme detection limits.
• Candidate detection (C1): A point-like source at projected ~1.1 au from α Cen A with SNR ~3 and elongation ~0.1″, repeatable across independent data subsets; brightness consistent with a 3–11 R⊕ planet (Neptune–Saturn-size) with 5–50% extra internal heating (Ag=0.3), or with a warm exozodiacal disk of ~60 zodis offset by ~0.3 au SW. Background star origin excluded by decade-prior pre-imaging. However, an unknown instrumental artifact cannot be ruled out.
• RV constraints: For α Cen A, RVs exclude Msini ≥53 M⊕ in the HZ, implying R≲7 R⊕ (using R⊕ M⊕^0.55) and compatible with C1’s allowed radii if planetary.

The NEAR experiment demonstrates that ground-based mid-infrared high-contrast imaging can achieve practical sensitivity to low-mass, temperate exoplanets in the habitable zones of the nearest Sun-like stars. By mitigating thermal background via DSM-based chopping and reducing starlight with a mid-IR AGPM coronagraph and shaped pupil, the campaign reached contrasts of ~2–3×10⁻⁶ at ~1″ around α Cen A, sufficient for sub-Neptune/Saturn-sized planets in thermal equilibrium with modest internal heating. Completeness analyses show high detection probabilities for Jovian-size planets and non-negligible sensitivity to smaller warm planets.

Scaling to Extremely Large Telescopes (ELTs) suggests that at 1″ separations the performance could be background-limited (SNR ∝ √t D², time ∝ D⁻⁴), yielding ~35 µJy (5σ, 1 h) for a NEAR-like instrument on a 39 m ELT—sufficient to detect an Earth analog (~20 µJy) near α Cen A in a few hours. Even if contrast-limited at 1″, speckle intensity scales ≲D⁻², implying achievable contrasts ~1.5×10⁻⁷ at 1″, again consistent with Earth-analog detectability. Further improvements (cold pupil stops to suppress AGPM glow, advanced apodizers, and non-common path aberration calibration) should enhance contrast-limited performance.

Regarding the candidate C1, its properties are consistent with either a warm Neptune-to-Saturn-sized planet or a localized warm exozodiacal dust structure, yet an instrumental origin cannot be excluded. Independent confirmation using additional mid-IR campaigns, RV, astrometry, or reflected-light imaging is required. The approach generalizes to other nearby stars (e.g., ε Eridani, ε Indi, τ Ceti), enabling broader habitable-zone surveys and setting the stage for ELT-era rocky planet imaging.

This work demonstrates a feasible ground-based mid-infrared strategy for directly imaging low-mass, temperate exoplanets around the nearest Sun-like stars. The NEAR campaign on VLT/VISIR achieved order-of-magnitude improvements in mass detection sensitivity compared to prior imaging limits, reaching ~3×10⁻⁶ contrast at ~1″ and high completeness for Jovian-size planets in α Centauri’s habitable zones, with sensitivity extending to sub-Neptune sizes for α Cen A. A tentative candidate (C1) may represent a Neptune–Saturn-size planet or exozodiacal dust but requires independent confirmation. The demonstrated techniques—mid-IR coronagraphy, shaped pupil, DSM-based AO and chopping, and robust post-processing—indicate that ELTs should be capable of detecting Earth analogs around α Cen A within hours. Future work should include repeated mid-IR observations to validate candidates, coordinated RV/astrometry/reflected-light monitoring, further instrument improvements (cold stops, apodizers, NCPA calibration), and extension of surveys to additional nearby systems.

• Significant systematic artifacts limited sensitivity and elevated false-positive rates in specific regions: detector persistence stripes from chopping and optical ghost arcs from the dichroic/filter. Although modeled and subtracted, residuals remain.
• Elevated local background near the coronagraph due to AGPM glow increased pixel noise (approximately doubled at ~1″) and reduced sensitivity near the inner working angle.
• Only 76.9 h of the ~100 h dataset met quality thresholds; remaining data were excluded due to high background, coronagraph misalignment, or AO issues.
• Detection thresholds correspond to SNR ~3; the candidate C1 is marginal and could be an unknown instrumental artifact, necessitating independent confirmation.
• Completeness is <100% even for large planets due to projection effects and masked regions (e.g., persistence stripes).