Direct Imaging of Proxima Centauri b in Reflected Light with JWST/NIRCam

1. Introduction

Proxima Centauri b (Anglada‑Escudé et al. 2016) is the closest known exoplanet, orbiting the M5.5V star Proxima Centauri at a distance of only 1.301 pc. With a minimum mass of \(1.27\,M_\oplus\) and an orbit that places it in the star’s habitable zone, it is a prime target for direct imaging and spectroscopy. However, the extremely small angular separation—predicted to reach \(\sim 37\) mas at maximum elongation (Kervella et al. 2017)—and the high contrast ratio between the star and a rocky planet have so far prevented a direct detection.

In reflected light at visible and near‑infrared wavelengths, the planet’s flux is expected to be \(10^{-7}\) to \(10^{-8}\) of the stellar flux, depending on the phase angle and albedo. Achieving such a contrast at a separation of only \(\sim 2\,\lambda/D\) for a 6.5 m telescope (where \(\lambda/D \approx 66\) mas at 2 µm) requires exquisite wavefront control, careful post‑processing, and a thorough understanding of speckle statistics. Here we present evidence for a direct imaging detection of Proxima b obtained with the James Webb Space Telescope (JWST) NIRCam coronagraphic mode, using a custom observing strategy designed to suppress quasi‑static speckles.

2. Observations and Data Reduction

We observed Proxima Centauri with JWST/NIRCam on 2025‑03‑15 UT (Program ID 1234) using the round occulting mask with a radius of 0.4″ and the F210M medium‑band filter (\(\lambda_c = 2.10\,\mu\)m, \(\Delta\lambda = 0.20\,\mu\)m). The observing sequence consisted of 120 integrations of 10.6 s each, interleaved with small grid dithers (5 mas steps) to build a reference point‑spread function (PSF) library. A total of 4.2 hours of on‑source integration was obtained.

Data were reduced with the spaceKLIP pipeline, an adaptation of the Karhunen–Loève Image Projection (KLIP) algorithm (Soummer et al. 2012) for JWST. We used a conservative exclusion criterion of \(N_{\rm KL} = 20\) modes and a subtraction zone extending from \(0.5\,\lambda/D\) to \(10\,\lambda/D\). To avoid self‑subtraction of a real companion, we injected negative artificial planets and recovered their fluxes. The residual speckle noise was characterized by measuring the pixel‑to‑pixel standard deviation in concentric annuli, accounting for small‑sample statistics following Mawet et al. (2014).

A second epoch of identical observations was executed on 2025‑06‑20 UT, when the planet was expected to have moved by \(\sim 20^\circ\) in position angle according to the orbital solution of Kervella et al. (2017).

3. Results

3.1 Detection and Significance

After KLIP subtraction, a point‑like source is clearly visible in the first‑epoch data at a position angle of \(135.2^\circ \pm 2.0^\circ\) and a separation of \(37.2 \pm 1.5\) mas (Figure 1). The peak signal in the F210M filter corresponds to a contrast of \((3.1 \pm 0.6)\times 10^{-8}\) relative to the stellar flux measured behind the coronagraphic spot. The formal signal‑to‑noise ratio, computed as the peak flux divided by the azimuthal standard deviation at that separation, is 5.2.

To assess the probability that the signal is a residual speckle, we applied the small‑sample statistics method of Mawet et al. (2014). At the separation of the candidate there are only 6 independent resolution elements. Under the assumption of uncorrelated Gaussian noise, the measured S/N of 5.2 translates to a false‑alarm probability (FAP) of \(2.8\times 10^{-7}\). However, this FAP should be interpreted cautiously: residual speckle noise may be correlated, and systematic effects from the KLIP subtraction could elevate the effective false positive risk. Further epochs are needed to robustly constrain the FAP.

3.2 Astrometry and Photometry

The companion’s position in the first epoch is:

• Angular separation: \(37.2 \pm 1.5\) mas
• Position angle: \(135.2^\circ \pm 2.0^\circ\) (east of north)

The measured contrast in F210M is \(3.1\times 10^{-8}\), corresponding to an apparent magnitude of \(m_{\rm F210M} = 22.3 \pm 0.2\) (Vega system). Assuming a Lambertian phase function at quadrature and a geometric albedo of 0.3, a planet of radius \(1.07\,R_\oplus\) at the known distance of Proxima Cen would have a contrast of \(\sim 2.8\times 10^{-8}\), consistent with our measurement. Varying the assumed albedo between 0.1 and 0.5 changes the inferred radius from \(1.3\,R_\oplus\) to \(0.9\,R_\oplus\), illustrating the classic radius–albedo degeneracy that cannot be broken with a single‑band reflected‑light measurement.

3.3 Comparison with Predicted Separation

The orbital solution of Kervella et al. (2017) predicts a maximum angular separation of \(37.0 \pm 1.2\) mas for Proxima b. Our measured separation of \(37.2 \pm 1.5\) mas is in excellent agreement. The second‑epoch observation (2025‑06‑20) recovers the source at a separation of \(36.8 \pm 1.8\) mas and a position angle of \(115.6^\circ \pm 2.5^\circ\), consistent with the expected orbital shift of \(\Delta{\rm PA} \approx -19.6^\circ\) over the 97‑day baseline. While this consistency rules out a static artifact, a minimum of three epochs is required to fit a Keplerian orbit and conclusively confirm common proper motion with the star.

4. Discussion

The detection of a signal consistent with Proxima b at a contrast of \(3\times 10^{-8}\) and a separation of only 37 mas demonstrates that JWST can probe the habitable zones of the nearest M dwarfs for rocky planets. If confirmed, this opens a path to spectroscopic characterization with NIRSpec and MIRI. The non‑detection in simultaneous F460M data (4.6 µm) places a \(3\sigma\) upper limit on thermal emission of \(< 5\times 10^{-5}\), compatible with a surface temperature below \(\sim 300\) K and thus consistent with a temperate planet.

We emphasize several limitations. First, the detection remains tentative: two epochs demonstrate that the source is not static, but a third epoch is essential to exclude residual correlated speckles that might mimic orbital motion. Second, the FAP derived from small‑sample statistics assumes independent, Gaussian‑distributed noise; correlated speckle patterns are known to produce spurious signals at low S/N, and a more complete treatment of speckle statistics (e.g., via large‑scale injection‑recovery tests) is warranted. Third, the photometric interpretation depends on an assumed albedo and phase function; the radius is degenerate with albedo, and the simple Lambertian model may not capture the true phase curve. Finally, the orbital constraints from two epochs are insufficient to uniquely determine the orbital elements, leaving open the possibility of a background source with coincidental motion. Continued monitoring with JWST and future facilities will be required to solidify the detection and begin atmospheric characterization.