Gravity Plus Adaptive Optics (GPAO)

• Gravity Plus Adaptive Optics (GPAO) is a high-order adaptive optics system that integrates visible and infrared wavefront sensing with both NGS and LGS modes to improve VLTI performance.
• It employs dual Shack–Hartmann sensors, advanced deformable mirrors, and SPARTA-based real-time control to achieve precise wavefront corrections and enhanced Strehl ratios.
• Laboratory tests and on-sky commissioning confirm GPAO’s capability to boost sensitivity, contrast, and fiber injection stability for demanding interferometric observations.

Gravity Plus Adaptive Optics (GPAO) is the dedicated high-order and laser-guide star adaptive optics system of the GRAVITY+ upgrade to the Very Large Telescope Interferometer (VLTI). It consists of four state-of-the-art AO systems equipping all 8 m-class Unit Telescopes (UTs) for the wavefront correction of the VLTI instruments, and it offers both visible and infrared Natural Guide Star (NGS) and Laser Guide Star (LGS) operations. In the published system descriptions, GPAO combines Shack–Hartmann wavefront sensing, an ALPAO kilo-class deformable mirror at M8, SPARTA-based real-time control, automated calibration of non-common-path aberrations, and dedicated alignment monitoring, with the stated goals of improving Strehl, faint-target sensitivity, sky coverage, contrast, and flux injection stability for interferometry (Collaboration, 25 Sep 2025).

1. Genealogy and position within the GRAVITY+ upgrade

GPAO emerged from the earlier GRAVITY adaptive-optics chain centered on CIAO, the GRAVITY Coudé Infrared Adaptive Optics system. CIAO was designed around a dedicated NIR Shack–Hartmann wavefront sensor in each UT Coudé laboratory, a 9×9 subaperture lenslet array, SPARTA-Light real-time control, and closed-loop correction at up to 500 Hz. In the design phase, the expected residual wavefront error was stated as $\leq 400\ \mathrm{nm}$ RMS, and simulated closed-loop performance for representative observing cases was reported in the range of $\sigma_{\rm WFE}\approx 363$–$522\ \mathrm{nm}$ depending on magnitude and geometry. In operation, CIAO routinely delivered K-band Strehl $\approx 0.6$–$0.7$ under median conditions and image jitter of 5–10 mas in K band, establishing the interferometric utility of infrared wavefront sensing for the Galactic Centre and related programs (Kendrew et al., 2012, Hippler et al., 2020).

The GRAVITY+ project broadened that architecture from an infrared, NGS-centered system to a facility-wide upgrade emphasizing all-sky, faint-science, and high-contrast near-infrared interferometry. The project description states that the upgrade includes wide-field off-axis fringe-tracking, new adaptive optics systems on all Unit Telescopes, and laser guide stars in an upgraded facility. At the project level, each UT was described as receiving a dedicated high-order AO module built around a Shack–Hartmann WFS, a 1353-actuator deformable mirror, and an ESO SPARTA real-time controller, together with a sodium LGS facility and low-order truth sensing. This transition marks the conceptual shift from CIAO as a GRAVITY-enabling subsystem to GPAO as the AO backbone of GRAVITY+ (Collaboration et al., 2023).

A later systems paper formalized four GPAO modes: IR-NGS, IR-LGS, VIS-NGS, and VIS-LGS. This mode structure is central to the upgrade because it couples high-order visible sensing, inherited infrared sensing, and LGS-based sky-coverage expansion within a single VLTI framework (Berdeu et al., 2024).

2. Optical architecture and observing modes

In the integrated GPAO design, each Wavefront Sensor unit on a UT comprises two Shack–Hartmann modules, NGS and LGS, mounted on a common opto-mechanical frame in the Coudé train downstream of the M8 deformable mirror and the M9 dichroic. After M8 and the M9 visible/IR split, a telecentric lens provides the patrol field, and two motorized XY-stages carry the NGS and LGS modules so that each can patrol the full Coudé field. The finalized WFS architecture reported after integration and characterization consists of a Natural Guide Star sensor with high-order 40×40 Shack–Hartmann sampling, a 4×4 low-order mode on the same NGS module, and a Laser Guide Star 30×30 sensor. The detector is a First Light OCAM2 EMCCD, operated from 100 Hz to 2 kHz, with effective read noise $<0.3\ \mathrm{e^-}$ RMS at gain 800 and 1 kHz (Bourdarot et al., 2024).

The beam routing is described in more detail in the first-light system paper. At the UT Nasmyth platform, the f/13 telescope beam is folded by the star-separator and dichroics. The M9 dichroic reflects $\lambda > 1\ \mu$m toward VLTI/STS and the CIAO-based NGS IR WFS, while transmitting 600–1000 nm to the visible WFS module. A telecentric f/47 lens below M9 reimages the patrol field onto two independent Shack–Hartmann WFS units on XY-translation stages. The visible NGS WFS is switchable between 40×40 high-order and 4×4 low-order lenslet arrays; the visible LGS WFS is a permanent 30×30 Shack–Hartmann fed by the sodium beacon; and the NGS IR WFS is the CIAO-heritage 9×9 Shack–Hartmann on a Saphira detector, reused unmodified inside GPAO (Collaboration, 25 Sep 2025).

Wavefront correction is applied by an ALPAO 43×43-actuator deformable mirror in the M8 position. Different papers describe it as an ALPAO 43×43 DM with 1432 active actuators, an ALPAO 43×43 DM with $\approx 1\,300$ actuators, or an ALPAO 41×41 actuator DM in early European tests; those descriptions correspond to different stages of the system definition and pre-series versus final units. In the integrated GPAO architecture, tip/tilt is off-loaded to M2 guiding, while the DM itself is also mounted on a gimbal or quasi-static mount for alignment functions (Millour et al., 4 Jun 2025, Bourdarot et al., 2024, Collaboration, 25 Sep 2025).

LGS operation is implemented with sodium lasers launched from each UT. The project paper described four continuous-wave, 20 W-class sodium-resonant lasers at $\lambda=589.159\ \mathrm{nm}$, one per UT, with UT1–UT3 side-launch and UT4 shared ELT-related launch optics. Later GPAO hardware descriptions specify a 20 W Raman-fibre sodium laser at 589.158 nm on the UT centre-piece, while the first-light paper presents the planned end-state as four 22 W sodium lasers and Laser Launch Telescopes per UT. Across these descriptions, the operational principle remains the same: LGS delivers high-order sensing, while a faint off-axis NGS provides low-order truth sensing for tip, tilt, and focus (Collaboration et al., 2023, Bourdarot et al., 2024, Collaboration, 25 Sep 2025).

3. Reconstruction, control, and performance modeling

The GPAO reconstruction formalism is presented in minimum-variance form. In the WFS paper, local slopes are collected into a vector $\boldsymbol{s}$ and modal coefficients or actuator commands into $\sigma_{\rm WFE}\approx 363$0, with reconstruction given by

$\sigma_{\rm WFE}\approx 363$1

where $\sigma_{\rm WFE}\approx 363$2 is the interaction matrix, $\sigma_{\rm WFE}\approx 363$3 the noise covariance, and $\sigma_{\rm WFE}\approx 363$4 a Tikhonov regularization parameter. GPAO employs a zonal DM-based reconstruction in closed loop, while modal control of up to 1,000 modes is available for bright NGS (Bourdarot et al., 2024).

The first-light system paper describes the operational control loop as implicitly modal on a Karhunen–Loève basis, including dark-hole KL modes and explicit low-order modes. In that implementation, high-order commands run at up to 2 kHz, routinely 1 kHz, while low-order pipelines run at 500 Hz. The control law is written

$\sigma_{\rm WFE}\approx 363$5

with $\sigma_{\rm WFE}\approx 363$6. Anti-windup and modal-leak paths are embedded in the high-order pipeline, and a pseudo-synthetic Interaction Matrix is refreshed every 5 s via a geometric model fit to measured mis-registration parameters, with numerical derotation of slopes to track DM/WFS pupil mis-rotation as the telescope slews (Collaboration, 25 Sep 2025).

At the level of fast performance prediction, GPAO is also described by a Maréchal-type error-budget model. For pure NGS operation, the simplified model writes

$\sigma_{\rm WFE}\approx 363$7

with

$\sigma_{\rm WFE}\approx 363$8

For LGS modes, the same formalism is extended to separate high-order and low-order loops and add a cone-effect term. The paper reports calibration of the NGS-mode coefficients against TIPTOP, an analytic-PSD long-exposure PSF simulator, and integration of the resulting model into the JMMC SearchFTT tool for observation planning (Berdeu et al., 2024).

Project-level descriptions give complementary control metrics. These include loop update rates of 500–1000 Hz in NGS mode and typically 500 Hz in LGS mode, total latency $\sigma_{\rm WFE}\approx 363$9 from WFS readout to DM command, and a closed-loop bandwidth of order 100 Hz in the earlier GRAVITY+ project description. The same paper decomposes the total residual phase variance as

$522\ \mathrm{nm}$0

explicitly retaining anisoplanatism and vibration terms in the LGS-enabled architecture (Collaboration et al., 2023).

4. Calibration, alignment, and laboratory qualification

A distinctive feature of GPAO is the degree to which integration and calibration were organized around dedicated laboratory infrastructure. The GRAVITY+ AO test bench at the Observatoire de la Côte d’Azur was built to replicate the optical and mechanical interfaces of an 8 m-class UT Coudé focus and Coudé room on a single 1.5 m × 2.4 m optical table enclosed in a 20 m² clean room. It delivered a telecentric F/46.7, 10 cm pupil to the Shack–Hartmann WFS, used an ALPAO DM241 with 41×41 actuators, and incorporated a rotating reflective phase screen to emulate $522\ \mathrm{nm}$1 turbulence, corresponding to 0.8″ seeing at 8 m aperture. The bench was designed to support full system integration and performance verification before Paranal deployment (Consortium et al., 2022).

The European qualification campaign in Nice extended this approach to end-to-end testing on a UT+atmosphere simulator. That bench reproduced a VLT UT Coudé focus and a turbulent Paranal atmosphere using two counter-rotating phase plates for 0.7″ seeing and additional plates for stress tests up to 1.4″ seeing and strong scintillation. Within this setup, GPAO validated fully automated non-common-path aberration calibration. The daytime procedure, implemented through the script gen_tec_ref_ncpa, sequentially probed approximately 40 Zernike or KL modes on an infrared Strehl camera at $522\ \mathrm{nm}$2m, converging in approximately 10 min to residual static aberrations $522\ \mathrm{nm}$3 RMS and raising the bench Strehl from $522\ \mathrm{nm}$4 to $522\ \mathrm{nm}$5 at 1.31 $522\ \mathrm{nm}$6m (Millour et al., 4 Jun 2025).

Alignment monitoring was another major calibration topic because GPAO contains moving optics between the DM and the WFS, including K-mirrors, atmospheric dispersion compensators, and field-patrol tables. The mis-registration papers define lateral errors $522\ \mathrm{nm}$7 as the sub-aperture-scale transverse shift between the DM actuator grid and the WFS sub-aperture grid, and they describe a two-stage strategy. In open loop, a perturbative 2D modal-correlation method uses a limited set of high-order DM modes and a Fourier-domain correlation map to estimate large shifts, with reported super-resolution of approximately 1/8 sub-aperture and convergence in 2–3 iterations. In closed loop, a non-perturbative telemetry-based Fourier estimator exploits the correlation between symmetric and anti-symmetric spatial-frequency components in the DM commands, enabling drift monitoring without perturbing the science stream. On the GPAO development bench, the open-loop method showed capture range up to at least $522\ \mathrm{nm}$8 and residual alignment better than $522\ \mathrm{nm}$9, while the closed-loop estimator corrected drifts to a few percent of a sub-aperture on timescales of seconds (Berdeu et al., 2024, Berdeu et al., 2024).

The European tests also stress-tested off-centered pupils, known DM actuator failures, induced field drift, and large derotator wobble. In those tests, the reported Strehl degradation remained below 5%, which the authors attribute to secondary controls such as mis-registration tracking, reference-slope updates, and modal leak integrators (Millour et al., 4 Jun 2025).

5. Quantitative performance

Laboratory and commissioning papers report GPAO performance in terms of Strehl, residual WFE, sensitivity, sky coverage, contrast, bandwidth, and flux injection stability. During integration and characterization of the WFS units, the internal wavefront error of the Shack–Hartmann chain was verified at $\approx 0.6$0 RMS, field vignetting was reported as none across the full patrol field, and laboratory closed-loop tests in NGS mode showed no degradation in Strehl up to 1,000 controlled modes, with a baseline of 500 modes at $\approx 0.6$1. In LGS mode, the reported baseline was 400 modes at $\approx 0.6$2, with no loss up to 500 modes (Bourdarot et al., 2024).

The Europe test campaign then reported end-to-end performance under median Paranal conditions $\approx 0.6$3. In NGS-VIS mode, using the 40×40 WFS at 1 kHz, GPAO reached $\approx 0.6$4 at $\approx 0.6$5 while controlling 500–800 modes. In LGS-VIS mode, using 30×30 high-order sensing plus 4×4 low-order sensing at 1 kHz/500 Hz, GPAO reached $\approx 0.6$6 at $\approx 0.6$7. The same paper states that measured Strehl-versus-magnitude curves overlaid the expected on-sky curves with 5–10% margin after baffling stray IR light in the NGS WFS. It also gives the Maréchal conversion from 1.31 $\approx 0.6$8m to K band,

$\approx 0.6$9

and notes that $0.7$0 at 2.2 $0.7$1m implies a residual WFE of approximately 100 nm RMS (Millour et al., 4 Jun 2025).

Project-level performance predictions provide the broader interferometric context. For bright NGS operation, the earlier GRAVITY+ paper states $0.7$2, typically 90%, in NGS mode for $0.7$3 mag and 0.8″ seeing. In LGS plus truth-sensor mode, it states $0.7$4–50% at $0.7$5 mag and 1.0″ seeing, with sky coverage increasing from approximately 10% at Galactic poles without LGS to approximately 80–90% with LGS plus an $0.7$6 mag NGS truth star within 1′. The same paper reports a factor 8–12 gain in fibre coupling into GRAVITY, a fringe-tracking limiting magnitude improvement from $0.7$7 to $0.7$8, coherent flux stability of approximately 20% peak-to-peak over 10 min, and a K-band sensitivity limit pushed to $0.7$9 mag for science targets with off-axis fringe guide up to 30″ (Collaboration et al., 2023).

The WFS paper adds science-facing performance estimates specific to GPAO. These include predicted Strehl up to 80% for bright NGS ($<0.3\ \mathrm{e^-}$0), 50% at $<0.3\ \mathrm{e^-}$1 in NGS mode, and 50% at $<0.3\ \mathrm{e^-}$2–19 in LGS mode under median seeing; residual WFE of 150–200 nm RMS; sky coverage increasing from approximately 10% for MACAO to $<0.3\ \mathrm{e^-}$3 with LGS plus NGS tip/tilt at $<0.3\ \mathrm{e^-}$4; raw contrast of approximately $<0.3\ \mathrm{e^-}$5 at $<0.3\ \mathrm{e^-}$6 mas in the NGS 40×40 high-contrast mode; and a 3.25 h, $<0.3\ \mathrm{e^-}$7 Galactic Centre detection limit improving from $<0.3\ \mathrm{e^-}$8 mag to $<0.3\ \mathrm{e^-}$9 mag (Bourdarot et al., 2024).

First-light NGS commissioning supplied the first on-sky GPAO metrics. Reported results include peak K-band Strehl of approximately 77% on $\lambda > 1\ \mu$0 in 0.4″ seeing and approximately 11% on $\lambda > 1\ \mu$1 in 1″ seeing, compared with a MACAO ceiling of approximately 40% even on bright stars. The RMS residual OPD after fringe tracking improved from approximately 150 nm to approximately 100 nm for $\lambda > 1\ \mu$2, the injection-stability metric $\lambda > 1\ \mu$3 increased from 0.31 to 0.62, and the rejection bandwidth at $\lambda > 1\ \mu$4 dB increased to approximately 45 Hz in VIS and 50 Hz in IR, versus 18 Hz with MACAO (Collaboration, 25 Sep 2025).

6. Scientific role, limitations, and broader adaptive-optics context

The scientific rationale of GPAO is explicitly tied to interferometric programs that are sensitivity-, contrast-, or sky-coverage-limited. The GRAVITY+ project paper identifies three major classes of use: black-hole mass measurements in active galactic nuclei across cosmic time, Galactic-centre observations that probe General Relativity through faint stars, and characterization of young exoplanets. These objectives are linked operationally to off-axis fringe tracking, higher Strehl, and improved fibre injection, rather than to image correction alone (Collaboration et al., 2023).

Science verification and first-light observations already illustrate that role. Reported early results include the first optical interferometry observations of a redshift $\lambda > 1\ \mu$5 quasar, spectroscopy of a cool brown dwarf with magnitude $\lambda > 1\ \mu$6, the first observations of a Class I young star with GRAVITY, and the first sub-micro arcsecond differential astrometry in the optical. More detailed commissioning examples include a $\lambda > 1\ \mu$7 quasar observed with G-Wide and GPAO NGS VIS, a $\lambda > 1\ \mu$8 brown dwarf observed in FAINT mode at $\lambda > 1\ \mu$9, HL Tau observed with NGS IR on XZ Tau and fringe-tracking on HL Tau, and $\approx 1\,300$0 Pic observed at $\approx 1\,300$1 with reported $\approx 1\,300$2as $\approx 1\,300$3 precision per channel across Br$\approx 1\,300$4 absorption (Collaboration, 25 Sep 2025).

Published modeling and test reports also state several limitations. The simplified Maréchal model lumps static non-common-path errors, telescope vibrations, and other non-modeled terms into an absolute floor $\approx 1\,300$5, and it notes that LGS-mode coefficients still require TIPTOP-based calibration once the laser modes are available. The same paper emphasizes that Maréchal validity is strict for $\approx 1\,300$6 rad and that on-sky tuning of coefficients remains necessary. The Nice bench campaign likewise states that it cannot recreate mesospheric sodium profile evolution or high-order LGS elongation, which are deferred to Paranal AIV (Berdeu et al., 2024, Millour et al., 4 Jun 2025).

Within the broader AO landscape, GPAO belongs to a class of systems that combine hybrid LGS/NGS sensing, statistical reconstruction, and mechanically constrained opto-mechanical calibration. A broader design lesson articulated in the MAORY literature is that gravity-invariant design principles, multi-conjugate DM architectures, and hybrid LGS/NGS sensing can be generalized to large-aperture telescopes, with the central objective of preserving alignment under gravity loads and using statistical control matrices to maximize final image quality. This does not make GPAO a multi-conjugate relay; rather, it situates GPAO within a wider progression from single-conjugate, instrument-specific AO toward tightly integrated interferometric and large-aperture AO systems (Ciliegi et al., 2021).