Annular Groove Phase Mask (AGPM) Overview

• AGPM is a subwavelength grating-based vector vortex coronagraph that creates a charge-2 azimuthal phase ramp for high-contrast imaging via destructive interference.
• It utilizes form birefringence in synthetic diamond and employs RCWA for precise design, ensuring broadband operation and a small inner working angle.
• Deployments on instruments like VLT/NACO and LBT/LMIRCam validate its performance in revealing close-in exoplanets and stellar companions.

The Annular Groove Phase Mask (AGPM) is a focal-plane vector vortex coronagraph implemented as a concentric subwavelength grating, typically etched in synthetic diamond, that synthesizes a charge-2 azimuthal phase ramp for high-contrast imaging at very small angular separations. In the language of vortex coronagraphy, it is the subwavelength grating vector vortex coronagraph of topological charge 2 (SGVC2). Its defining combination of small inner working angle, broadband operation through form birefringence and geometric phase, high throughput, and a clear $360^\circ$ discovery space made it the first mature subwavelength-grating vortex architecture widely deployed on thermal-infrared astronomical instruments (Absil et al., 2014).

1. Definition, nomenclature, and optical principle

The AGPM consists of concentric annular grooves that realize a space-variant half-wave plate. Its local fast-axis orientation rotates with azimuth, so that circular polarization components acquire opposite geometric phases and the transmitted field receives the vortex phase ramp required for coronagraphy. In standard notation, the focal-plane phase function is

$\Phi(\theta)=l\,\theta,$

with complex transmission

$V(\theta)=e^{i l\theta},$

and $l=2$ for the AGPM. Over a full revolution, the phase therefore grows from $0$ to $4\pi$ (Absil et al., 2014).

The device is a vector vortex rather than a scalar spiral phase plate. Its achromatization is based on the geometric, or Pancharatnam-Berry, phase. For an even-charge vortex, the local fast-axis orientation obeys

$\psi(\theta)=\frac{l}{2}\theta,$

so the imparted geometric phase is $2\psi=l\theta$. In AGPM implementations, this rotating half-wave plate is synthesized by a subwavelength grating whose form birefringence provides the required retardance (Delacroix et al., 2014).

In coronagraphic operation, the azimuthal phase ramp redistributes on-axis stellar light outside the geometric pupil in the downstream pupil plane, where a Lyot stop blocks it. Off-axis sources do not satisfy the same symmetry about the mask center and are transmitted. This continuous phase ramp distinguishes the AGPM from earlier phase-mask architectures based on discrete phase steps. One consequence is the absence of quadrant transitions and the retention of a full $360^\circ$ discovery space (Delacroix et al., 2014).

A useful historical nuance is that some broad reviews of astronomical phase masks discuss Roddier masks, Lyot nulling, and dual-zone phase masks without explicitly treating AGPMs or vortex masks. In that broader taxonomy, the AGPM belongs conceptually to focal-plane phase-mask coronographs that tailor the stellar electric field to induce destructive interference downstream, but its polarization-engineered geometric-phase implementation lies outside those specific historical treatments (Dohlen, 2018).

2. Electromagnetic design and coronagraphic behavior

The AGPM relies on subwavelength-grating form birefringence. To remain in the zeroth-order regime and avoid higher diffraction orders, the grating period must satisfy

$p<\frac{\lambda}{n}.$

The birefringent retardance of the grating is designed to approximate half-wave operation across the target band,

$\Phi(\theta)=l\,\theta,$0

where $\Phi(\theta)=l\,\theta,$1 is the groove depth and $\Phi(\theta)=l\,\theta,$2 is the effective-index difference produced by the subwavelength structure (Absil et al., 2014).

Within this framework, the principal design variables are period, depth, fill factor or duty cycle, and sidewall angle. Because $\Phi(\theta)=l\,\theta,$3 depends on all of them, AGPM design is typically performed with Rigorous Coupled Wave Analysis (RCWA), often using measured geometry from scanning electron microscopy to predict actual performance rather than nominal performance (Absil et al., 2014).

For a charge-2 AGPM, the reported inner working angle is about $\Phi(\theta)=l\,\theta,$4, and the device is routinely described as operating “down to the diffraction limit of the telescope” (Absil et al., 2014). Charge increase improves robustness at the price of a larger inner working angle: for charge 4, the reported inner working angle is approximately $\Phi(\theta)=l\,\theta,$5 (Delacroix et al., 2014).

The literature summarized here also shows that care is required when reading “off-axis transmission” scalings. Some AGPM studies state that near-axis transmission for a charge-2 vortex scales approximately as $\Phi(\theta)=l\,\theta,$6 (Delacroix et al., 2014), and equivalent statements appear in early on-sky AGPM reporting (Mawet et al., 2013). A more recent MMT/MIRAC-5 upgrade note writes the scaling as

$\Phi(\theta)=l\,\theta,$7

which gives quartic behavior for $\Phi(\theta)=l\,\theta,$8 (Miller et al., 5 Aug 2025). This suggests that the exact reported law depends on the definition being used, such as field transmission, intensity transmission, or an instrument-specific effective scaling, and that the underlying convention should be checked before comparing published values.

Even topological charge is required for the classical coronagraphic effect. Higher charges are being investigated because leakage from finite stellar diameter and pointing jitter scales favorably with charge. In the ELT-oriented SGVC4 analysis, residual stellar-disk leakage is described as scaling as $\Phi(\theta)=l\,\theta,$9, making higher-charge vortices attractive for partially resolved stars in the near-infrared, albeit at the cost of larger inner working angle and more difficult grating design (Delacroix et al., 2014).

3. Materials, fabrication, metrology, and representative geometries

Synthetic diamond is the canonical AGPM substrate. The stated rationale includes wide transparency from the ultraviolet to microwaves, high refractive index, toughness, hardness, high thermal conductivity, electrical insulation, chemical stability, low thermal expansion, and compatibility with durable high-aspect-ratio microstructures in the thermal infrared (Absil et al., 2014). In mid-infrared AGPM work, the diamond refractive index is quoted as $V(\theta)=e^{i l\theta},$0 from $V(\theta)=e^{i l\theta},$1 to $V(\theta)=e^{i l\theta},$2 (Delacroix et al., 2014).

The microfabrication route matured from the first full AGPM attempts in 2009 to science-grade devices by 2012 after control of etch depth and sidewall angle was achieved. Reported processes include nano-imprint lithography and reactive ion etching for mid-infrared prototypes, inductively coupled plasma reactive ion etching for L-band devices, and solvent-assisted micromolding for improved pattern transfer in later L-band optimization campaigns (Delacroix et al., 2014). Uppsala University’s Ångström Laboratory is specifically identified with the development of deep plasma etching into diamond (Absil et al., 2014).

Metrology is central because AGPM performance is extremely sensitive to small geometric errors. Scanning electron microscopy of cleaved cross-sections is repeatedly used to retrieve period, groove depth, fill factor, and sidewall angle, and the resulting measured geometry is fed back into RCWA. In later L-band optimization work, tolerances of order $V(\theta)=e^{i l\theta},$3 nm in top width, $V(\theta)=e^{i l\theta},$4 nm in depth, and $V(\theta)=e^{i l\theta},$5 in sidewall angle are identified as performance-critical (Catalan et al., 2016).

The back side of the diamond is commonly patterned with an anti-reflective grating to suppress multiple reflections and ghosts. In L band, backside reflectance was reduced from about $V(\theta)=e^{i l\theta},$6 to about $V(\theta)=e^{i l\theta},$7, and total transmission of an optimized component was measured at approximately $V(\theta)=e^{i l\theta},$8 across the band (Delacroix et al., 2013). For the VISIR upper-N implementation, finished components including the anti-reflective grating reached total transmittance between $V(\theta)=e^{i l\theta},$9 and $l=2$0 over $l=2$1–$l=2$2 (Delacroix et al., 2014).

Representative geometries reported in the AGPM literature include the following. In L band, one optimized family uses $l=2$3, with groove depths around $l=2$4, top widths near $l=2$5–$l=2$6, and sidewall angles near $l=2$7–$l=2$8 (Delacroix et al., 2013). For the VISIR upper-N AGPM, the reported robust solution is $l=2$9, $0$0, $0$1, and $0$2, with top line width $0$3 (Delacroix et al., 2014).

A recurring fabrication issue is that non-vertical sidewalls become increasingly problematic as periods shrink toward shorter wavelengths. This is explicit in K-band development, where target periods near $0$4 nm introduce redeposition and non-vertical sidewall challenges (Absil et al., 2014).

4. Laboratory validation and measured performance

By 2013–2014, mid-infrared AGPMs had progressed from design prediction to quantitative laboratory demonstration. On the YACADIRE coronagraphic bench, an L-band AGPM delivered a broadband raw null depth of $0$5 over $0$6–$0$7, corresponding to a raw contrast of about $0$8 at $0$9. The result matched projections from RCWA when measured grating parameters and ghost terms were included (Delacroix et al., 2013).

The VORTEX project summary reports broadband peak rejection up to $4\pi$0 in the L band, which translates into a raw contrast of about $4\pi$1 at $4\pi$2 (Absil et al., 2014). Subsequent L-band optimization based on controlled re-etching improved starlight rejection to $4\pi$3 in a broadband L filter, corresponding to a raw contrast of about $4\pi$4 at two resolution elements from the star for a perfect input wave front on a circular, unobstructed aperture (Catalan et al., 2016).

Bench results also exposed dominant sensitivities. Focus errors of $4\pi$5 mm at $4\pi$6 degraded null depth by a factor of about $4\pi$7 in the 2013 L-band laboratory demonstration, corresponding to $4\pi$8 nm peak-to-valley, or about $4\pi$9 nm RMS, defocus aberration (Delacroix et al., 2013). Later bench work treated low-order aberration sensing as a system-level requirement rather than a device-level afterthought, leading to the VODCA concept of a reflective Lyot stop sending rejected starlight to a second camera for low-order wavefront sensing, coupled to a deformable mirror for correction (Jolivet et al., 2016).

For N band, early AGPM studies emphasized that full cryogenic characterization lagged L-band validation. Nonetheless, RCWA based on measured parameters predicted peak rejections of at least a few tens for N-band components, and preliminary internal-source tests on VISIR were reported as consistent with those expectations (Absil et al., 2014). A later N-band upgrade study for MIRAC-5/MMT reported monochromatic rejection values measured on the CEA Saclay bench of $\psi(\theta)=\frac{l}{2}\theta,$0 at $\psi(\theta)=\frac{l}{2}\theta,$1, $\psi(\theta)=\frac{l}{2}\theta,$2 at $\psi(\theta)=\frac{l}{2}\theta,$3, $\psi(\theta)=\frac{l}{2}\theta,$4 at $\psi(\theta)=\frac{l}{2}\theta,$5, $\psi(\theta)=\frac{l}{2}\theta,$6 at $\psi(\theta)=\frac{l}{2}\theta,$7, and $\psi(\theta)=\frac{l}{2}\theta,$8 at $\psi(\theta)=\frac{l}{2}\theta,$9, demonstrating both strong peak performance and substantial chromatic roll-off across the usable band (Miller et al., 5 Aug 2025).

An important practical conclusion of the laboratory literature is that once rejection approaches the $2\psi=l\theta$0 level in L band, the intrinsic mask is no longer necessarily the dominant limitation. Low-order aberrations, alignment, ghosting, and bench noise floors become comparably important (Catalan et al., 2016).

5. Instrument deployments and early scientific results

Science-grade AGPMs were installed on VLT/NACO, VLT/VISIR, and LBT/LMIRCam in 2012–2013, and later on Keck/NIRC2 in 2015 (Absil et al., 2014). These deployments established AGPMs as operational coronagraphs rather than solely laboratory devices.

At VLT/NACO, first light in December 2012 yielded an instantaneous on-sky peak rejection of about $2\psi=l\theta$1. During these first observations, a late-type stellar companion to HD 4691 was discovered at only two beamwidths from the star. The reported separation was $2\psi=l\theta$2 arcsec, with $2\psi=l\theta$3, and the companion was interpreted as an M2V star of approximately $2\psi=l\theta$4 and $2\psi=l\theta$5 K at a projected separation of $2\psi=l\theta$6 AU (Mawet et al., 2013).

The same NACO AGPM delivered science verification on $2\psi=l\theta$7 Pictoris on 31 January 2013. Over a 3.5-hour sequence, the observations achieved unprecedented sensitivity down to the diffraction limit, approximately $2\psi=l\theta$8, enabling searches down to about $2\psi=l\theta$9 AU separations and providing new constraints on debris-disk morphology at small angles (Absil et al., 2014).

At LBT/LMIRCam, first light in October 2013 produced new L-band images of HR 8799. During the first commissioning hours, all four known planets were clearly detected at high SNR, and the system reached sensitivity down to the diffraction limit of $360^\circ$0, corresponding to about $360^\circ$1 AU at $360^\circ$2 pc (Defrère et al., 2014). Although the achieved rejection in that first run was only about $360^\circ$3, the study explicitly attributed the shortfall to focus and centering limitations rather than intrinsic AGPM performance, and expected improvement to at least $360^\circ$4 with better optimization (Defrère et al., 2014).

The longer “three years of harvest” synthesis adds instrument-level comparison. LMIRCam achieved the deepest contrasts among the reported thermal-IR vortex modes, attributed to low thermal background, excellent LBTAO performance with Strehl up to $360^\circ$5, and an optimized Lyot stop. NACO and NIRC2 gave similar contrasts, with NIRC2 benefiting from a deeper background limit in the quoted comparison (Absil et al., 2016).

VISIR’s AGPM path was more protracted because commissioning was interrupted by detector issues, but internal-source tests indicated rejection of at least $360^\circ$6, and the AGPM was integrated with a custom “12_4_AGP” filter spanning $360^\circ$7–$360^\circ$8 and a dedicated Lyot stop (Absil et al., 2016).

6. System-level limitations, centering, and data reduction

AGPM performance is strongly coupled to the optical system around it. Several leakage channels are explicitly quantified in the thermal-infrared on-sky literature. Diffraction by a central obstruction contributes a leakage fraction scaling approximately as $360^\circ$9, where $p<\frac{\lambda}{n}.$0 is the obstruction radius and $p<\frac{\lambda}{n}.$1 the pupil radius. For $p<\frac{\lambda}{n}.$2, the quoted leakage is about $p<\frac{\lambda}{n}.$3. Pointing jitter contributes

$p<\frac{\lambda}{n}.$4

and higher-order aberration leakage is approximated by

$p<\frac{\lambda}{n}.$5

with $p<\frac{\lambda}{n}.$6 the Strehl ratio (Absil et al., 2016).

These expressions explain why intrinsic mask chromatic leakage, often measured below $p<\frac{\lambda}{n}.$7 on the bench, is not the dominant term in many on-sky datasets (Absil et al., 2016). They also motivate several operational developments. The VORTEX project explored tip-tilt estimation from science-camera images, reflective Lyot stops for low-order sensing, and centering cross hairs printed on some AGPMs (Absil et al., 2014). For the VISIR AGPM, centering fiducials were deposited by Aerosol Jet Printing as $p<\frac{\lambda}{n}.$8-wide silver lines, with cryogenic robustness verified by interferometric profilometry (Delacroix et al., 2014).

A distinctive operational feature of thermal-IR AGPMs placed upstream of a cold stop is the appearance of a bright diffuse spot at the vortex center on a uniform sky background. This is attributed to incoherent thermal emission from outside the geometric pupil that the vortex partially diffracts into the pupil image. The effect disappears if the AGPM is placed downstream of a cold stop, but in practice it proved useful for locating the vortex center and was removed by background subtraction or post-processing (Absil et al., 2016).

Closed-loop pointing control became a major practical advance. QACITS, adapted from earlier phase-mask work to vortex operations, was commissioned at Keck/NIRC2 and achieved closed-loop pointing accuracy down to about $p<\frac{\lambda}{n}.$9 RMS, approximately $\Phi(\theta)=l\,\theta,$00 mas at Keck (Absil et al., 2016). A later MIRAC-5/MMT program plans QACITS residuals below $\Phi(\theta)=l\,\theta,$01–$\Phi(\theta)=l\,\theta,$02 at $\Phi(\theta)=l\,\theta,$03 (Miller et al., 5 Aug 2025).

Post-processing is equally consequential. PCA-ADI pipelines were already used in the early $\Phi(\theta)=l\,\theta,$04 Pictoris and HR 8799 AGPM reductions (Absil et al., 2014). The NaCo-centered CenteR pipeline later showed that improved centering alone can materially change detectability: in VLT/NaCo $\Phi(\theta)=l\,\theta,$05-band AGPM data, the S/N of companions around $\Phi(\theta)=l\,\theta,$06 Pictoris and R CrA improved by $\Phi(\theta)=l\,\theta,$07 and $\Phi(\theta)=l\,\theta,$08, respectively, relative to other state-of-the-art reductions (Godoy et al., 2021). That study recommended centering frames on the AGPM position for PCA construction, but derotating and stacking around the stellar position.

7. Extensions, redesigns, and current directions

Several AGPM development lines follow directly from the trade-offs of the charge-2 architecture. The first is higher topological charge. A perfectly continuous rotationally symmetric binary grating for $\Phi(\theta)=l\,\theta,$09 cannot remain subwavelength everywhere, so the SGVC4 program proposed an optimized curved-line discretization with azimuthally varying period and constant filling factor. RCWA predicted mean null depths $\Phi(\theta)=l\,\theta,$10, corresponding to broadband rejection $\Phi(\theta)=l\,\theta,$11, across both K and L bands for the optimized design, while preserving the intended phase ramp more faithfully than earlier straight-slice discretizations (Delacroix et al., 2014).

The second line is spectral downscaling. K-band AGPM development targets periods near $\Phi(\theta)=l\,\theta,$12 nm, but introduces new difficulties from redeposition and sidewall control (Absil et al., 2014). This is consistent with the general trend that AGPM fabrication becomes more difficult as the period shrinks.

The third line is inverse and surrogate-assisted design of the central singular region, where the annular geometry departs from strict periodicity and full-wave simulation is required. A 2023 study introduced a data-efficient surrogate optimization framework combining a deep U-Net surrogate with particle swarm optimization. For the annular-groove center, direct optimization achieved a leakage metric $\Phi(\theta)=l\,\theta,$13, while the surrogate workflow found $\Phi(\theta)=l\,\theta,$14 with roughly four times fewer full-wave evaluations; overall, the method reduced simulations by up to $\Phi(\theta)=l\,\theta,$15 compared to conventional optimization techniques (Roy et al., 2023). The same study found that annular grooves outperformed nanofin centers in absolute leakage, and that high-performing nanofin solutions tended to self-organize into ring-like arrangements (Roy et al., 2023).

A fourth line is system-level integration for ELTs and new mid-infrared facilities. ELT/METIS studies optimized diamond AGPMs for L, M, and split N bands, with mean theoretical null depths of $\Phi(\theta)=l\,\theta,$16 in L, $\Phi(\theta)=l\,\theta,$17 in M, $\Phi(\theta)=l\,\theta,$18 in lower N, and $\Phi(\theta)=l\,\theta,$19 in upper N, under ideal-wavefront assumptions (Carlomagno et al., 2016). More recently, the MIRAC-5/MMT MAPS coronagraphic upgrade adopted an N-band AGPM with an optimized Lyot stop computed in HEEPS and projected off-axis throughput of about $\Phi(\theta)=l\,\theta,$20, illustrating the migration of AGPM practice into newer adaptive-optics mid-infrared platforms (Miller et al., 5 Aug 2025).

Taken together, these developments show that the AGPM is no longer only a fixed component design. It has become a family of related vortex architectures spanning device physics, microfabrication, Lyot-stop co-design, low-order sensing, inverse electromagnetic optimization, and ELT-specific robustness engineering.