The final design of GMagAO-X: high-contrast imaging at first-light of the GMT

GMagAO-X will be the first-light high-contrast imager on the 25 m Giant Magellan Telescope. The driving science case for GMagAO-X is characterization of the atmospheres of nearby rocky exoplanets such as Proxima Centauri b. The revolutionary increase in spatial resolution and sensitivity provided by GMagAO-X will enable detailed study of such planets for the first time. Additional science cases include: reflected light characterization of mature giant planets; measurement of young extrasolar giant planet variability; characterization of circumstellar disks at unprecedented spatial resolution; characterization of benchmark stellar atmospheres at high spectral resolution; and mapping of resolved objects such as giant stars and asteroids. These, and many more, science cases will be enabled by a 21,000 actuator extreme adaptive optics (ExAO) system, an integrated coronagraphic wavefront control system with dedicated deformable mirrors, and a suite of imagers and spectrographs. Science-driven performance requirements for GMagAO-X include achieving a Strehl ratio of 70% at 800 nm on 8th mag and brighter stars, and exoplanet characterization at planet:star flux-ratios of 1e-7 at 4 lambda/D (26 mas at 800 nm) separation. GMagAO-X has been added to the GMT project baseline plan and is in the final design phase, aiming to complete FDR in March, 2027. The instrument is on track to be ready at first-light of the GMT in the mid 2030s. We provide a brief update of the instrument designed to achieve our ambitious performance targets.

Key milestones in the GMagAO-X project to date include: February, 2021: Conceptual design started; September, 2021: Conceptual Design Review passed; December, 2021: Preliminary design started; February, 2024: Preliminary Design Review passed; September, 2025: GMagAO-X added to GMT baseline, final design started; September, 2027: Planned Final Design Review.

The key instrument parameters of GMagAO-X include: Wavelength Coverage requirement 600–1900 nm, goal 450–1900 nm, stretch goal 350–1900 nm; Spatial Resolution requirement 4.9 mas, goal 3.7 mas, stretch goal 2.8 mas; Spectral Resolution requirement BB: 10% bands, Low: 20–100, High: 1000–65,000, goal 100,000; Guide Star I Magnitude Range requirement-1.5–13, goal-1.5–15; Field of View requirement 3” x 3”, goal 3.5” x 6”; Coronagraph Contrast requirement 1e-7 @ 4λ/D, goal 1e-7 @ 2λ/D and 1e-8 @ 6λ/D, stretch goal 1e-8 @ 1λ/D and 1e-9 @ 5λ/D.

GMagAO-X is intended to occupy a Folded Port (FP) on the Gregorian Instrument Rotator (GIR) of the GMT. It consists of a two-level optical table, which will be air-floated to provided vibration isolation. The GIR rotates and tips with elevation of the telescope, and as such the FPs are not gravity invariant. To enable the gravity invariance required for air isolation, GMagAO-X is designed to counter rotate. When GMagAO-X is in operation the GIR will be locked so that GMagAO-X will be oriented along the elevation axis of the GMT. GMagAO-X will also supply a dedicated custom tertiary mirror (M3). After M3, a relay includes a fast-steering mirror (FSM) and then couples the beam to the gravity invariant main section of the instrument. After the fore-optics, the beam passes through a pupil alignment periscope, a k-mirror field rotator to keep the pupil alignment fixed, and an atmospheric dispersion compensator (ADC). A “woofer” deformable mirror (DM) is used to provide large-stroke for correcting low-order aberrations at lower speeds. At PDR a 3228 actuator DM was planned for the “woofer”, however we are currently reevaluating the trade off between number of actuators, actuator stroke, and temporal response. After the woofer, the beam is relayed to the 21,000 “tweeter“ DM which provides the high-speed and high-spatial-frequency control. A key lesson-learned from on-sky operation of MagAO-X is that stable high-Strehl wavefronts are required for achieving the highest performance from the coronagraph WFS and control (WFS&C) system. To achieve a projected actuator pitch similar to MagAO-X (∼14 cm), ∼3000 illuminated actuators are needed per segment. GMagAO-X will achieve this with a unique “Parallel DM” architecture, consisting of seven 3000 actuator MEMS deformable mirrors. A hexagonal prism with its 6 segments coated as mirrors splits the GMT aperture such that each segment is imaged onto a 3K MEMS DM. The central hole passes the center segment to its DM in the back. The parallel DM has been prototyped and demonstrated, albeit without MEMS installed, in the High Contrast Adaptive Optics phasing Testbed (HCAT) at the University of Arizona. After the “tweeter”, the beam is split between the wavefront sensors (WFS) and the coronagraph. The coronagraph beam is relayed to the upper level and to a 3000 actuator non-common path correcting (NCPC) DM, an architecture which has been demonstrated on-sky with MagAO-X. The coronagraph is a Lyot-style architecture which will support a range of designs, including the Phase-Apodized Pupil Lyot Knife-Edge Coronagraph (PAPLKEC), and the Phase Induced Amplitude Apodization (PIAA) coronagraph with transmissive complex focal plane masks. Focal plane instrumentation such as imagers covering the optical through 1 µm, and near-IR through H band as well as spectrographs are included. For highest efficiency we are designing an on-board lower-resolution IFU. It may be possible to couple GMagAO-X to the facility G-CLEF and GMTNIRS spectrographs as a fiber fed integral field unit (IFU).

The required stroke for an AO system is proportional to (D/r0)5/6. MagAO-X has 3.5 µm of total surface stroke in its “tweeter”, and 15 µm of tip/tilt stroke in its woofer. Moving from a 6.5 m to a 25.4 m telescope means that up to a factor of 3 more stroke is needed (mitigated somewhat by the outer scale). The FSM in the fore-optics of GMagAO-X alleviates the most demanding tip/tilt stroke requirements. For higher orders, we are planning to use a MEMS device with at least 5.5 µm of surface stroke. This has necessitated a development effort in collaboration with Boston Micromachines Corporation (BMC) to improve the temporal response time of these devices. The results show that the “Multi-5.5-Dash-2” architecture is nearly as fast as the existing MagAO-X device and meets the GMagAO-X specification of < 100µs rise/fall time.

The lower-bench contains both a Holographic Dispersed Fringe Sensor (HDFS) for coarse phasing control and a 3-sided pyramid WFS (PyWFS) for higher order control. We plan to have both visible and near-IR (through 1.8 µm) PyWFS. Within the coronagraph, both focal-plane low-order WFS (FLOWFS) and Lyot-plane LOWFS (LLOWFS) will make use of rejected light. To obtain high contrast we will employ focal-plane WFS techniques, such as implicit Electric Field Conjugation (iEFC). The self-coherent camera (SCC) is also under consideration. The software system for GMagAO-X is based on the proven software suite developed for MagAO-X, now called the “eXtreme Wavefront Control Toolkit” (XWCTk). The XWCTk is based on the Compute and Control for Adaptive Optics (CACAO) system. CACAO is in routine use at SCExAO and on MagAO-X. The XWCTk will be adapted to interface with the GMT software and controls system, ensuring that GMagAO-X functions safely and efficiently as part of the observatory system without losing the proven heritage of MagAO-X on-sky operations.

GMagAO-X will be the first-light ExAO coronagraph on the GMT. It represents the earliest opportunity of the ELT era to begin search nearby terrestrial planets for life. GMagAO-X is now in the GMT project plan and is in the final design phase. The scientific potential of GMagAO-X motivates a positive decision to move forward with construction of the GMT as rapidly as possible.

Improvements to AI Systems:

1. Adaptive Optics Control Optimization

• Implement AI-based predictive wavefront control using the GMagAO-X architecture (parallel DMs, woofer/tweeter split, FSM).

• Train reinforcement learning models to optimize DM stroke allocation and temporal response, reducing latency and improving Strehl ratio stability under varying atmospheric turbulence.

• Use neural networks to model the nonlinear relationship between actuator commands and wavefront phase, enabling faster convergence than traditional matrix-based controllers.

2. Coronagraphic Wavefront Sensing & Control (WFS&C)

• Develop AI algorithms for implicit Electric Field Conjugation (iEFC) and self-coherent camera (SCC) processing, using deep learning to estimate and correct residual speckle noise in real time.

• Train convolutional neural networks (CNNs) on simulated and on-sky data to predict optimal DM shapes for contrast levels down to 1e-8 at 2λ/D, improving exoplanet detection sensitivity.

3. Multi-Object Spectral Classification

• Build a transformer-based model to process high-resolution spectra (R=100,000) from GMagAO-X’s IFU, automatically classifying exoplanet atmospheres (e.g., Proxima Cen b) by detecting molecular signatures (H2O, CO, CH4) and clouds.

• Enable unsupervised clustering of spectral features to identify novel atmospheric compositions without prior labels.

4. Real-Time Data Reduction & Denoising

• Use generative adversarial networks (GANs) to denoise low-signal coronagraphic images, recovering planet signals at flux ratios of 1e-7 with 4λ/D separation.

• Implement autoencoders for on-the-fly calibration of non-common path aberrations, reducing systematic errors in Strehl ratio and astrometric precision.

5. Autonomous Observing & Scheduling

• Create a reinforcement learning agent to manage GMagAO-X’s counter-rotation, ADC, and field rotator, optimizing pointing and gravity-invariant alignment during long exposures.

• Use AI to prioritize science targets (e.g., rocky exoplanets vs. disks) based on real-time atmospheric conditions and instrument status, maximizing scientific yield.

6. Telescope Phasing & Segment Control

• Train a deep learning model on Holographic Dispersed Fringe Sensor (HDFS) data to predict piston/tip/tilt errors across GMT’s 7 segments, enabling faster and more robust coarse phasing than current algorithms.

• Implement a neural network to fuse HDFS and pyramid WFS data for joint fine phasing, reducing the time to reach diffraction-limited performance.

7. Simulation-to-Reality Transfer

• Develop a digital twin of GMagAO-X using physics-informed neural networks, trained on the HCAT testbed data, to simulate wavefront control and coronagraphic performance.

• Use domain adaptation to transfer AI models trained in simulation to on-sky operations, accelerating commissioning and reducing risk.

---

What the Improved AI System Can Do:

• Achieve stable 70% Strehl at 800 nm on 8th mag stars by predicting and correcting atmospheric turbulence faster than current AO loops.

• Detect and characterize exoplanets at flux ratios of 1e-7 at 4λ/D, and potentially 1e-8 at 2λ/D, enabling atmospheric study of Proxima Cen b.

• Automatically classify exoplanet atmospheres from spectral data, identifying biosignatures (e.g., O2, CH4) with high confidence.

• Reduce data processing time from hours to minutes, enabling real-time target-of-opportunity observations.

• Self-calibrate and adapt to changing telescope conditions, minimizing downtime and human intervention.

• Optimize observing schedules to capture transient events (e.g., stellar flares, disk variability) while maintaining high-contrast performance.

• Provide robust phasing control for the GMT’s segmented mirror, ensuring diffraction-limited imaging across the full 25 m aperture.