Directly imaging exoplanets is challenging because quasi-static phase aberrations in the pupil plane ( speckles ) can mimic the signal of a companion at small angular separations .
Kernel phase , which is a generalization of closure phase ( known from sparse aperture masking ) , is independent of pupil plane phase noise to second order and allows for a robust calibration of full pupil , extreme adaptive optics observations .
We applied kernel phase combined with a principal component based calibration process to a suitable but not optimal , high cadence , pupil stabilized L ’ band ( 3.8 ~ { } \text { \textmu m } ) data set from the ESO archive .
We detect eight low-mass companions , five of which were previously unknown , and two have angular separations of \sim 0.8 – 1.2 ~ { } \lambda / D ( i.e . \sim 80 – 110 ~ { } \text { mas } ) , demonstrating that kernel phase achieves a resolution below the classical diffraction limit of a telescope .
While we reach a 5 \sigma contrast limit of \sim 1 / 100 at such angular separations , we demonstrate that an optimized observing strategy with more diversity of PSF references ( e.g .
star-hopping sequences ) would have led to a better calibration and even better performance .
As such , kernel phase is a promising technique for achieving the best possible resolution with future space-based telescopes ( e.g .
JWST ) , which are limited by the mirror size rather than atmospheric turbulence , and with a dedicated calibration process also for extreme adaptive optics facilities from the ground .