A Partial Lyman Limit Absorber in the Halo of a Galaxy Pair: A Possible Signature of Gas Inflow

We present an analysis of a partial Lyman limit system at z = 0.87641 detected in the HST /COS spectrum of the background quasar LBQS 0107−0235. The absorber exhibits a simple kinematic structure, with the metal-lines and the H I Lyman-series absorption well described by a single component. Photoionization modeling yields a gas metallicity of one-tenth solar and a hydrogen number density of nH ≈ 8.5×10−4 cm−3 (log10 (nH /cm−3) ≈ −3.1). At the absorber redshift, the V LT /MUSE data show two galaxies (G1 and G2) at normalized impact parameters of ρ/Rvir ≈ 0.9 and velocity separations of ∆v = 18 and 99 km s−1, respectively, from the absorber. Both galaxies have rotating disks with stellar masses of M∗ ≈ 6 × 109 and ≈ 2.2 × 1010 M⊙. Their 100-Myr-averaged star formation rates are ≈ 2.5 and ≈ 2.2 M⊙ yr−1, though their instantaneous rates place G2 on the star-forming main sequence and G1 above it, which is actively star-forming at this redshift. The absorber is positioned very close to the projected major axis of both galaxies. The absorber’s orientation, kinematics, and sub-solar metallicity (log10 (Z/Z⊙) = −1.05) are consistent with the absorption tracing a sub-solar metallicity inflowing stream, though a galaxy–galaxy interaction origin cannot be excluded. We discuss these scenarios in the context of cosmological simulations of cold-mode accretion and CGM gas flows around galaxies with halos of mass Mh ≲ 1012 M⊙.

Improvements to AI Systems:

1. Multiwavelength Data Fusion for Gas–Galaxy Association

• Improvement: Develop an AI model that jointly analyzes HST/COS absorption spectra (H I Lyman series, metal lines) and VLT/MUSE integral-field spectroscopy (galaxy kinematics, stellar masses, SFRs) to automatically identify and rank candidate host galaxies for intergalactic absorbers.

• Capability: The system can predict the most probable galaxy counterpart(s) for a given absorber by integrating velocity offsets, impact parameters (ρ/Rvir), and disk orientation angles, reducing manual cross-matching errors.

2. Physical Parameter Inference from Photoionization Modeling

• Improvement: Train a neural network (e.g., normalizing flows or Bayesian neural networks) on grids of photoionization models (e.g., Cloudy) to map observed absorption-line ratios (e.g., C IV/H I, Si IV/H I) directly to posterior distributions of gas metallicity (Z), hydrogen number density (nH), and ionization parameter.

• Capability: The AI can rapidly infer sub-solar metallicity (e.g., log10(Z/Z⊙) = −1.05) and nH ≈ 8.5×10−4 cm−3 from noisy spectra, with uncertainty quantification, without requiring manual iterative fitting.

3. Kinematic Classification of CGM Gas Flows

• Improvement: Implement a supervised classifier (e.g., gradient-boosted trees or a small transformer) trained on simulated cold-mode accretion and galaxy–galaxy interaction outputs (from cosmological simulations like TNG or EAGLE) to label observed absorbers as inflowing stream, outflow, or tidal debris based on kinematic structure (single vs. multi-component), metallicity, and galaxy-relative velocity.

• Capability: Given an absorber’s velocity separation (e.g., ∆v = 18 and 99 km/s) and proximity to galaxy major axes, the system can output a probability score for each origin scenario, aiding interpretation of CGM gas flows.

4. Automated Detection of Sub-Solar Metallicity Inflow Signatures

• Improvement: Build a deep-learning anomaly detector that scans large quasar spectra for absorption systems with simple kinematic structure (single-component H I and metal lines) and low metallicity, flagging them as candidate cold-accretion streams.

• Capability: The system can sift through thousands of spectra to identify rare, low-metallicity absorbers aligned with galaxy disks, enabling statistical studies of cold-mode accretion without human visual inspection.

5. Simulation-to-Observation Transfer Learning for CGM Models

• Improvement: Use domain adaptation to train an AI model on synthetic absorption spectra generated from cosmological simulations (with known inflow/outflow labels) and then fine-tune it on real HST/COS data to predict gas flow origins, accounting for instrumental resolution and noise.

• Capability: The improved system can generalize from simulated halos (Mh ≲ 1012 M⊙) to observed systems like LBQS 0107−0235, providing physically grounded classifications even when galaxy–galaxy interaction scenarios are ambiguous.

6. Real-Time Galaxy–Absorber Geometry Estimator

• Improvement: Create a geometric reasoning module (e.g., graph neural network) that ingests galaxy disk orientations (position angle, inclination) and absorber sky position to compute the projected major-axis offset and normalized impact parameter (ρ/Rvir) automatically.

• Capability: The AI can instantly output whether an absorber lies along a galaxy’s major axis (as in this case), which is a key geometric indicator for inflowing streams, enabling rapid triage of future CGM surveys.