Kernel Phase Interferometry (KPI)

• Kernel Phase Interferometry is a technique that models telescope pupils as virtual interferometric arrays, using kernel phases to cancel first-order phase errors.
• It employs singular value decomposition to derive a kernel operator that isolates intrinsic astrophysical signals while mitigating instrumental and atmospheric perturbations.
• KPI enhances angular resolution and contrast, enabling direct detection of close companions and detailed studies of stellar systems with both space- and ground-based telescopes.

Kernel Phase Interferometry (KPI) is a post-processing analysis technique that enables high-contrast, super-resolution imaging within and even below the classical diffraction limit of a telescope. By mathematically modeling a filled or partially obstructed pupil as a virtual interferometric array of subapertures, KPI generalizes the principles of closure phase from sparse aperture masking to arbitrary, potentially redundant, telescope pupils. The technique forms specially constructed linear combinations—kernel phases—of Fourier-plane image phases such that, to first order, all contributions from small residual instrumental and atmospheric wavefront errors are nullified. KPI thus provides imaging observables that retain intrinsic astrophysical phase signatures while suppressing first-order instrumental aberrations. This capability markedly enhances the angular resolution and contrast achievable with both space- and ground-based telescopes, particularly for the direct detection of substellar companions at separations well inside the nominal $\lambda/D$ diffraction limit (Adelman et al., 15 Oct 2025, Kammerer et al., 2022, Sivaramakrishnan et al., 2022, Pope et al., 2013, Martinache, 2010, Martinache, 2013).

1. Mathematical Foundations of Kernel Phase Interferometry

KPI operates on the foundational observation that—under the small-aberration, high-Strehl regime—the phase measured in the Fourier plane of a direct image is a linear function of the underlying pupil-plane wavefront errors. Consider a pupil discretized into $N$ subapertures: let $\varphi$ be the vector of phase errors, and denote the complex visibility phase at each of the $M$ distinct spatial frequencies sampled by the pupil as $\boldsymbol{\Phi}$. The measurement equation can be written as

$$\boldsymbol{\Phi} = \boldsymbol{\Phi}_0 + R^{-1}A\varphi,$$

where $\boldsymbol{\Phi}_0$ contains the object-intrinsic phases, $A$ is the $M\times N$ phase-transfer (baseline) matrix encoding the mapping from pupil to Fourier plane, and $R$ is a diagonal redundancy matrix accounting for multiple (u,v) contributions (Adelman et al., 15 Oct 2025, Martinache, 2011, Kammerer et al., 2022).

Kernel phases $N$0 are built by finding a matrix $N$1 spanning the left null space of $N$2: $N$3. Left-multiplying the measurement equation by $N$4 eliminates all terms depending on the instrumental phase,

$N$5

so that, to first order, $N$6 is only sensitive to intrinsic object structure (e.g., binarity or disk asymmetry). In the particular case of a non-redundant sparse mask, kernel phases reduce to traditional closure phases; for redundant or filled apertures, they generalize closure relations to the full null space of $N$7. The kernel operator $N$8 is constructed efficiently via singular value decomposition (SVD) of $N$9, yielding $\varphi$0 independent kernel phases per configuration (Sivaramakrishnan et al., 2022, Adelman et al., 15 Oct 2025, Pope et al., 2013).

Error propagation and covariance calculation follow by

$\varphi$1

allowing for rigorous statistical detection and model fitting (Sivaramakrishnan et al., 2022, Ceau et al., 2019). These properties underpin the robustness and precision of KPI observables.

2. Algorithmic Workflow and Pipeline Implementation

KPI analysis begins with precise modeling of the telescope pupil, including obscurations and support structures. The pupil is discretized on a Cartesian, hexagonal, or custom grid, the pitch of which ($\varphi$2) is tuned to sample the spatial frequencies of interest. Each pair of subapertures defines a baseline, and their ensemble forms the baseline mapping matrix $\varphi$3, with baseline redundancies accounted for in $\varphi$4 (Adelman et al., 15 Oct 2025, Martinache et al., 2020, Kammerer et al., 2022).

• Image Preprocessing: Input images (after standard calibrations) are background-subtracted and cropped to retain all baselines. Bad pixels are detected and replaced (median filtering or Fourier-plane inpainting). Super-Gaussian windows suppress edge noise (Adelman et al., 15 Oct 2025, Kammerer et al., 2022).
• Phase Extraction: The preprocessed image is Fourier transformed. Complex visibilities at the sampled (u,v) points are extracted, with the argument yielding raw Fourier phases $\varphi$5.
• Kernel Phase Calculation: Application of the SVD-derived operator $\varphi$6 to the measured phases produces the kernel phase vector $\varphi$7.
• Calibration: Depending on available data, calibration may involve subtracting reference-star kernel phases (RDI), angular differential kernel phase (ADKP), or principal component projection (KLIP) to further remove residual systematics (Adelman et al., 15 Oct 2025, Laugier et al., 2020, Kammerer et al., 2022).
• Data Structures and Exchange: Standardization is achieved via the KPFITS format, which contains pupil models, baseline mapping, kernel matrices, kernel phase data, and associated metadata, supporting robust data exchange and reproducibility (Kammerer et al., 2022).
• Model Fitting and Detection: Grid and gradient-based searches, likelihood maximization, and MCMC posterior sampling are used to fit binary or parametric models directly to $\varphi$8, with significance metrics derived from rigorous $\varphi$9 statistics and hypothesis testing theory (Kammerer et al., 2022, Ceau et al., 2019).

3. Performance Benchmarks and Detection Limits

KPI consistently demonstrates superior resolution and contrast at small separations compared to conventional image-plane or PSF fitting methods. Key quantitative results include:

Instrument	Band	$M$0 (mas)	5$M$1 Contrast at $M$2	IWA ($M$3mas)	Reference
JWST/MIRI	7.7$M$4m	244	$M$5mag 5–6	$M$6244	(Adelman et al., 15 Oct 2025)
JWST/NIRISS	4.8$M$7m	189	$M$86.5 mag @200mas	70	(Kammerer et al., 2022, Sivaramakrishnan et al., 2022)
HST/NICMOS NIC1	1.1$M$9m	95	$\boldsymbol{\Phi}$0100:1 @200mas	50	(Pope et al., 2013, Factor et al., 2022)
Palomar/PHARO	2.1$\boldsymbol{\Phi}$1m	100	30:1 @130mas	50–100	(Pope et al., 2015)
SCExAO/CHARIS	2.2$\boldsymbol{\Phi}$2m	70	$\boldsymbol{\Phi}$3m 7–8 @40mas	40	(Chaushev et al., 2024, Chaushev et al., 2023)

KPI’s inner working angle (IWA) routinely reaches or even slightly exceeds the classical Rayleigh limit, with super-resolution routinely achieved ($\boldsymbol{\Phi}$4 or less), especially under high Strehl, space-based, or extreme AO conditions (Adelman et al., 15 Oct 2025, Sivaramakrishnan et al., 2022, Martinache, 2011, Factor et al., 2022). Contrast limits are generally photon-noise dominated at larger angles, with systematics and residual correlation noise becoming limiting factors at the smallest separations—particularly in the presence of uncalibrated or poorly modeled instrumental aberrations.

Sophisticated detection limit estimation leverages statistically independent kernel observables and Neyman–Pearson or likelihood-ratio hypothesis testing, yielding reproducible false-alarm rates and confidence intervals on fitted parameters (Ceau et al., 2019, N'Diaye et al., 2022).

4. Calibration Strategies and Extensions

KPI’s principal robustness derives from first-order insensitivity to pupil-plane phase errors. However, systematic leakage due to imperfect pupil modeling, uncalibrated amplitude errors, and subtle higher-order effects remains a calibration challenge (Martinache et al., 2020, Pope, 2016). Current best practices include:

• Reference-Star Differential Imaging (RDI): Calibration through contemporaneous, spatially co-located point sources mitigates systematic bias, particularly in space-based campaigns (Adelman et al., 15 Oct 2025, Kammerer et al., 2022).
• Angular Differential Kernel Phases (ADKP): Observations at multiple field rotations, with biases projected out via linear algebra, allow the science target itself to serve as its own calibrator, reducing reliance on external calibrators and improving residual distributions (Laugier et al., 2020).
• Spectral Differential Imaging (SDI): Multi-wavelength IFS KPI exploits the relative constancy of systematics across adjacent channels to self-calibrate and isolate spectral-line signals (e.g., in searches for Br-γ accreting protoplanets) (Chaushev et al., 2023, Chaushev et al., 2024).
• Principal Component (KLIP) Calibration: Projecting onto the orthogonal complement of the leading calibrator principal components further suppresses systematic leakage (Kammerer et al., 2022).

Accurate, amplitude-weighted pupil modeling (“grey” models) and precise redundancy accountancy are critical for minimizing calibration-induced detection limit biases (Martinache et al., 2020).

5. Applications and Scientific Results

KPI has been applied to a variety of high-impact science cases:

• Substellar and Exoplanet Companion Detection: Recovery and discovery of tight (sub-AU to 10 AU) binaries and planetary companions in space (HST/NICMOS; (Pope et al., 2013, Factor et al., 2022)), JWST/NIRISS (Kammerer et al., 2022, Sivaramakrishnan et al., 2022) and ground-based AO (Palomar/PHARO, SCExAO/CHARIS) datasets.
• Post-main-sequence Exoplanet Surveys: JWST/MIRI KPI enables detection of inward-migrated giant planets around white dwarfs at separations $\boldsymbol{\Phi}$5 AU, probing orbital regions inaccessible to other methods (Adelman et al., 15 Oct 2025).
• Protoplanet and Disk Asymmetry Searches: KPI with IFS datasets (CHARIS) leverages both continuum and line emission (e.g., Br-γ) to search for accreting protoplanets at 4–16 AU in young stellar systems, setting competitive limits on accretion rates and substellar masses (Chaushev et al., 2024, Chaushev et al., 2023).
• Population Studies and Binary Fraction Determinations: Large archival sweeps provide unbiased companion statistics at separations and contrasts previously unattainable, e.g., increased ultracool dwarf binary fractions (Pope et al., 2013, Factor et al., 2022).

The high angular resolution and throughput realized by KPI enable the direct imaging of faint companions and structures close to bright stars, producing precise relative astrometry and photometry.

6. Current Limitations, Systematic Effects, and Future Directions

The dominant limiting factors for KPI performance are accurate and stable pupil modeling, sufficient AO correction or naturally high wavefront stability (as in space), and robust calibration. Systematic errors arising from incomplete pupil transmission models, orientation misalignments, amplitude aberrations, and temporal drift in systematics can set the detection floor—especially at the smallest inner working angles (Martinache et al., 2020, Pope, 2016, Pope et al., 2015). Strategies for further improvement include:

• Comprehensive amplitude (kernel amplitude) calibration: Extending kernel-phase analysis to amplitude observables to jointly calibrate throughput and scintillation errors (Pope, 2016).
• Next-generation ELTs: Application to thirty- and forty-meter class telescopes, where KPI can achieve $\boldsymbol{\Phi}$65–10 mas resolving power in the near-IR, subject to adequate Strehl and careful mask or pupil design (Martinache, 2013).
• Photon-noise–limited performance: Achievable through extreme AO plus long integrations and advanced multi-channel calibration (e.g., SDI, ADKP), enabling detection of companions at contrasts $\boldsymbol{\Phi}$7 within $\boldsymbol{\Phi}$8 (Kammerer et al., 2022, Ceau et al., 2019, Chaushev et al., 2023).
• Hybridization: Combining kernel-phase observables with classical and machine learning–based high-contrast imaging techniques for joint suppression of phase and amplitude systematics.

KPI continues to be extended to new instrumental platforms, including JWST/NIRSpec and MIRI, as well as ground-based ELTs with advanced AO.

7. Summary and Scientific Significance

Kernel Phase Interferometry constitutes a transformative development in high-contrast imaging, delivering robust, self-calibrating phase observables that enable direct detection of faint companions, circumstellar structure, and disk asymmetries within and beyond the diffraction limit of both space-based and ground-based telescopes. By recasting direct imaging as generalized interferometry, KPI leverages modern linear algebra and statistical detection theory, providing a rigorously quantifiable path to super-resolution imaging, dynamical mass measurement, and comprehensive population studies. Its versatility, photon throughput, and effectiveness at the smallest angular separations establish KPI as a central technique for the current and future eras of exoplanet and binary star science (Adelman et al., 15 Oct 2025, Kammerer et al., 2022, Sivaramakrishnan et al., 2022, Factor et al., 2022, Martinache et al., 2020, Pope et al., 2015).