Dust production in the harsh environment of Sgr A* - MIRI/JWST observation of the O-rich asymptotic giant branch star IRS 3

This paper presents MIRI/JWST observations of the O-rich AGB star IRS 3, located about 0.17 parsecs in projection from Sgr A* in the Galactic Center. The study aims to constrain the dust composition and line-emitting species within the envelope of IRS 3.

Observations and Data Reduction: IRS 3 was observed with the MIRI Medium Resolution Spectrometer (MRS) on 21st April 2025 as part of the guaranteed time observations (GTO) program Mid-Infrared Characterisation of Nearby Iconic galaxy Centres (MICONIC). The observations covered a total wavelength range from 4.9 to 27.9 µm. The data was affected by strong saturation in several bands, which was remediated with a custom reduction for single-group data. Partial spectral discontinuities due to overlapping channels were corrected with polynomials.

Extinction Correction: The spectral analysis used a stellar-based mid-infrared extinction curve from Kemper et al. (2004), normalized to the continuum-dominated emission of IRS 3 in the spectral range of 5.0-7.7 µm. A foreground factor of Afg = 0.3 mag was set, translating into a multiplicative factor of 3.66 at 9.7 µm. The results were compared with extinction laws from Fritz et al. (2011) and von Fellenberg et al. (2025), and the overall morphology of the spectrum was found to be robust against the choice of foreground correction.

Results: After correcting for foreground dust extinction, the spectrum shows a broad 9.7 µm silicate stretching feature accompanied by an O-Si-O bending mode absorption feature at 18.5 µm. The optical depth of the 9.7 µm absorption feature is τ9.7 = 2.98 ± 0.07, and the O-Si-O bending mode at 18.5 µm yields an optical depth of τ18.5 = 0.85 ± 0.01. The resulting ratio between τ9.7 and τ18.5 is 3.5 ± 0.1, which classifies IRS 3 as an O-rich star. The paper states: The observed spectral shape and radiative transfer models support the classification of IRS 3 as an O-rich AGB star.

Radiative Transfer Modeling: Using the 3D Monte-Carlo radiation transfer code Hyperion, the observed spectrum was modeled with a best-fit stellar luminosity of 60000 L⊙. The model shows a multi-shell configuration with a significant temperature gradient between the inner and outer layers of the envelope. The envelope is characterized by a shell-like distribution with a temperature gradient of ≈ 1000 K. A combination of alumina (Al2O3) and amorphous silicates accurately reproduces the observed spectrum. The model includes an inner region with a temperature of 1200 K, and three shells with temperatures of 280-300 K, 180 K, and 80-100 K. The characteristic radii of these shells are approximately 949 AU, 2214 AU, and 6325 AU, respectively, with a total envelope radius of about 104 AU.

Water Detection: For the first time, clear signs of water (H2O) were found in the envelope of IRS 3. Using HITRAN transition line lists, a synthetic spectrum was constructed and fitted to the region around 6.0 µm. The H2O component was modeled with a temperature of 700 K and column densities of 1.5 × 1017 cm−2. The paper notes: This finding demonstrates that the harsh radiation-dominated environment of Sgr A* does not inhibit dust formation or the survival of molecular species such as H2 O.

Stellar Parameters: Based on the radiative transfer analysis and stellar tracks using PARSEC-COLIBRI isochrones, the paper infers a stellar mass of about 6 M⊙ and a corresponding age of ≈ 72 Myr. The effective temperature is 2800 K, and the metallicity is assumed to be +0.35 dex.

Mass-Loss Rate: Assuming a wind velocity of vw = 15 km/s, the mass-loss rate is estimated to be 6 × 10−5 M⊙ yr−1. Using the measured stand-off distance of the bow shock (R0 ≈ 4316 AU), the ambient number density at the bow-shock apex is estimated to be nH ≈ 4.5 × 102 cm−3, with a range of nH ≈ (2.3 × 102 − 1.2 × 103) cm−3 considering statistical uncertainty.

Conclusions: The key findings are summarized as follows:

1. For the first time, a continuous mid-infrared spectrum of IRS 3 is observed.

2. The morphology of the spectrum is robust against the choice of the extinction law.

3. IRS 3 is an O-rich AGB star with a stellar mass of ≈ 6 M⊙ and an age of ≈72 Myr, potentially formed in the Nuclear Stellar Cluster.

4. Using radiative-transfer models with an outflow-driven multiple shell setup, the dereddened spectrum of IRS 3 is reconstructed.

5. With a best-fit stellar luminosity of 60000 L⊙, a mass-loss rate of about 6 × 10−5 M⊙ yr−1 is estimated with an assumed wind velocity of vw = 15 km/s.

6. For the first time, clear signs of H2O are found in the envelope of IRS 3 in the inner parsec of the Milky Way that resist the harsh conditions in the vicinity of Sgr A*.

Improvements to AI Systems:

1. Robust Saturation Remediation for Space-Based Spectrometers:

• Improvement: Train a deep-learning model to automatically detect and correct saturated single-group data in MIRI/MRS observations, using the custom reduction approach as a training set.

• Capability: The AI can autonomously reduce saturated JWST spectra in real-time, preserving spectral fidelity across overlapping channels without manual polynomial corrections.

2. Extinction-Law-Agnostic Spectral Morphology Classification:

• Improvement: Develop a classifier that extracts invariant spectral features (e.g., silicate feature ratios, continuum shape) robust to foreground extinction choices, trained on IRS 3 and synthetic spectra with varying extinction laws.

• Capability: The AI can classify evolved stars (O-rich vs. C-rich) from mid-IR spectra even when the extinction law is unknown or debated, improving reliability in high-extinction regions like the Galactic Center.

3. Automated Radiative Transfer Model Fitting with Multi-Shell Envelopes:

• Improvement: Implement a neural network surrogate for the Hyperion 3D Monte-Carlo code to rapidly fit multi-shell dust temperature gradients, luminosity, and composition (e.g., alumina + amorphous silicates) to observed spectra.

• Capability: The AI can infer stellar luminosity, envelope shell radii, and temperature profiles in minutes instead of days, enabling large-scale surveys of AGB stars in crowded, dusty environments.

4. Molecular Line Identification in Noisy, Extincted Spectra:

• Improvement: Train a transformer-based model on HITRAN line lists and synthetic spectra to detect weak molecular features (e.g., H2O at 6.0 µm) in heavily extincted, low-S/N mid-IR data, with uncertainty quantification on column density and temperature.

• Capability: The AI can automatically flag and characterize water, CO2, or other molecules in circumstellar envelopes even when features are partially masked by silicate absorption, expanding molecular inventories in extreme environments.

5. Stellar Parameter and Age Inference from Spectral + Isochrone Data:

• Improvement: Create a Bayesian neural network that combines radiative-transfer-derived luminosity and effective temperature with PARSEC-COLIBRI isochrones to jointly estimate stellar mass, age, and metallicity, including systematic uncertainties from extinction and wind velocity.

• Capability: The AI can produce probabilistic stellar evolutionary histories for individual stars in the Galactic Center, enabling population synthesis studies of the Nuclear Stellar Cluster with quantified confidence.

6. Bow-Shock and Ambient Density Estimator from Stand-Off Distance:

• Improvement: Develop a physics-informed regression model that inputs bow-shock stand-off distance, stellar wind velocity, and mass-loss rate to predict ambient hydrogen number density, with error propagation from all input uncertainties.

• Capability: The AI can map the local interstellar medium density around evolved stars in high-radiation environments, aiding studies of feedback and star formation near Sgr A*.

What the Improved AI System Can Do:

• Automatically process and analyze JWST MIRI spectra of dusty, extincted stars, from raw saturated data to physical parameter inference.

• Provide robust classification and molecular detection in extreme galactic environments without manual extinction-law assumptions.

• Rapidly generate and validate radiative transfer models for hundreds of AGB stars, enabling statistical studies of mass loss and dust formation in the Galactic Center.

• Deliver probabilistic stellar ages, masses, and local ISM densities, directly supporting models of stellar evolution and galactic chemical enrichment in the presence of a supermassive black hole.