PASSAGES: The Discovery of a Strongly Lensed Protocluster Core Candidate at Cosmic Noon

Summary

This paper reports the discovery of J0846+15 (J0846), one of the most gas- and dust-rich protocluster core candidates, identified from the Planck All-Sky Survey to Analyze Gravitationally lensed Extreme Starbursts (PASSAGES) sample. The system consists of a foreground galaxy cluster (J0846.FG) at z = 0.77 that strongly lenses a background system (J0846.BG) of at least 11 dusty star-forming galaxies (DSFGs) at z = 2.660–2.669.

Key observational findings:

• ALMA Band 3 observations (Cycle 5, program 2017.1.01214.S) uncovered 18 CO(3–2) emission-line detections, some of which are multiply-imaged systems, all within a redshift range of z = 2.660–2.669 (velocity range ∆V ≈ 800 km s−1).

• The 18 CO images originate from 11 unique sources: four are multiply-imaged (ID1, ID4, ID6, ID7) and seven are singly-imaged (ID2, ID3, ID5, ID8, ID9, ID10, ID11).

• Three additional multiply-imaged systems (ID12, ID13, ID14) were identified in HST/Gemini imaging without ALMA counterparts, with geometric lens model predicted redshifts of z ≈ 2.77–2.90.

• The foreground cluster J0846.FG has nine spectroscopically confirmed member galaxies within z = 0.750–0.772, with a mean redshift of z = 0.766 ± 0.002 (biweight estimator).

Lens model:

• A parametric lens model was constructed using GLAFIC, constrained by the positions of multiply-imaged systems and singly-imaged galaxies (with penalty terms to avoid predicting multiple images).

• The best-fit model reproduces the angular positions of multiply-imaged systems to an rms separation of 0.14″.

• Magnification factors for the CO images range from µ ≃ 1.5 to 25, with a mean magnification of 2 over the entire region.

• The source plane reconstruction shows the 11 galaxies are contained within a projected physical extent of 280 × 150 kpc, with a velocity dispersion of σv = 246 ± 72 km s−1.

Physical properties:

• The total apparent (uncorrected for lensing) star formation rate is µSFR = 39900+23000−12900 M⊙ yr−1, derived from a modified blackbody fit to the unresolved dust SED (Td = 47.1+8·8−6·4 K, Md = 2.21+0·62−0·53 × 1010 M⊙, β = 1.63+0·19−0·14, A = 482+331−291 kpc2).

• The intrinsic (magnification-corrected) SFR is estimated to be SFR = 5200+3200−2000 M⊙ yr−1, using a magnification factor of µ = 7.7 ± 1.5 based on the ratio of total intrinsic to observed CO line luminosities.

• The total molecular gas mass (using αCO(3–2) 1 M⊙ (K km s−1 pc2)−1) is MISM = (2.0 ± 0.3) × 1011 M⊙.

• The dynamical mass, estimated via the virial theorem (Mdyn = 3Rσ2/G) using R = 280 kpc and σv = 246 ± 72 km s−1, is Mdyn = 1.2 ± 0.6 × 1013 M⊙.

• Individual dust masses range from 0.21 × 108 to 4.2 × 108 M⊙, with molecular gas-to-dust mass ratios of 90–700.

Galaxy population diversity:

• Six of 11 galaxies (ID2, ID3, ID4, ID5, ID7, ID9) exhibit kinematic features consistent with ordered disk rotation (smooth, monotonic velocity gradients).

• Two of 11 systems (ID6, ID8) show disturbed kinematics indicative of mergers/tidal interactions, with ID8.2 having a FWHM of 786 ± 105 km s−1 and ID6 showing a tidal bridge between components.

• Three systems remain undetermined in their kinematic classification.

Evolutionary context:

• The system is compared to other DSFG-rich protocluster cores (e.g., SPT2349-56 with 23 galaxies at z = 4.304, total SFR 4,480 M⊙ yr−1), making J0846.BG among the most starbursting protocluster core candidates reported.

• Based on Chiang et al. (2013) evolutionary predictions, the estimated halo mass of Mdyn = 1.2 ± 0.6 × 1013 M⊙ may evolve into a ≳2 × 1014 M⊙ cluster at z = 0.

• The molecular gas depletion timescale is measured to be < 50 Myr, though this could be longer depending on αCO assumptions and potential gas replenishment.

Additional notes:

• The foreground cluster shows evidence of a merger scenario: a radio wide-angle tail (WAT) galaxy (tentative BCG), a straight arc (ID7.2a/b) typical of merging mass components, and a possible bimodal velocity distribution.

• A bright foreground elliptical at z = 0.357 (a line-of-sight perturber) was modeled as an additional NFW profile with a dynamical mass prior from MUSE line width measurements.

• The paper acknowledges that the full extent of J0846.BG may be larger than the arcmin FOV, and that more than 50 additional lensed sources are photometrically identified in the HST F160W image that could be associated with the structure.

Based on the scientific paper, here are specific improvements that can be made to AI systems, along with what the improved systems can do:

• Improvement: Train a deep learning model (e.g., convolutional neural networks or transformer-based vision models) on ALMA CO(3–2) moment-0 maps and HST F160W images to automatically identify multiply-imaged galaxies, including merging image pairs (e.g., ID1ab, ID4ab, ID6.1ab) and counterimages (e.g., ID1c, ID4c, ID6.1c).

• What the improved AI can do:

• Detect and label all 18 CO detections and 8 non-CO image systems (ID12–14) without manual inspection.

• Distinguish between single-image and multiple-image systems based on morphology, parity flips, and proximity to critical curves.

• Predict the number of unique sources (11) from the observed 18 images, reducing human bias in source deblending.

• Improvement: Build a supervised classifier using moment-1 velocity maps and line profiles (FWHM, velocity gradients) to categorize galaxies into ordered disks, mergers, or undetermined systems.

• What the improved AI can do:

• Automatically classify the 11 member galaxies (e.g., 6 disks, 2 mergers, 3 undetermined) with quantified confidence.

• Detect non-monotonic velocity gradients and multiple velocity components (e.g., ID8.2 with FWHM 786 km/s) that indicate mergers or tidal interactions.

• Flag galaxies with symmetric velocity structures across critical curves (e.g., ID6.1ab) as strong lensing candidates.

• Improvement: Replace manual MCMC parameter sampling with a neural network-based emulator or normalizing flow to predict lens model parameters (NFW halo, pseudo-Jaffe profiles, perturber mass) from image positions and redshifts.

• What the improved AI can do:

• Reduce computational time for posterior sampling from thousands of chains to seconds.

• Provide robust uncertainty estimates on magnification factors (µ) and geometric redshifts (e.g., z geo for ID12–14) that account for systematic model degeneracies (e.g., mass-sheet degeneracy).

• Automatically identify which image systems provide the strongest constraints on the critical curve and caustic shape.

• Improvement: Train a generative model (e.g., variational autoencoder or diffusion model) to de-lens observed images back to the source plane, preserving intrinsic morphology and kinematics.

• What the improved AI can do:

• Reconstruct the 11 member galaxies in the source plane (280 × 150 kpc) with minimal distortion, even for highly magnified systems (µ up to 80 for ID6ab).

• Automatically measure intrinsic sizes, velocity gradients, and merger features (e.g., tidal bridges between ID6.1 and ID6.2) without manual aperture selection.

• Predict the intrinsic SFR and gas masses per galaxy, correcting for differential magnification.

• Improvement: Develop a cross-matching algorithm that combines ALMA CO(3–2), 3mm continuum, VLA 6 GHz, HST F160W, and Gemini r′/z′ data to identify counterparts and measure fluxes automatically.

• What the improved AI can do:

• Match 8/18 CO detections to 3mm continuum and 5/18 to 6 GHz radio emission, flagging non-detections.

• Measure photometric redshifts and colors for >50 additional lensed sources in the HST FOV, potentially identifying new protocluster members.

• Automatically compute dust masses and gas-to-dust ratios for each member using modified blackbody fits.

• Improvement: Train a random forest or gradient boosting model on simulated protoclusters (e.g., from Chiang et al. 2013) to classify observed systems as protocluster cores based on velocity dispersion, physical extent, SFR, and gas mass.

• What the improved AI can do:

• Automatically classify J0846.BG as a protocluster core candidate with a probability score, given σ v = 246 ± 72 km/s, extent 280 × 150 kpc, and SFR 5200 M⊙/yr.

• Predict the likely descendant halo mass at z=0 (e.g., 2 × 10 14 M⊙) and the expected stellar mass buildup over 1 Gyr.

• Flag systems with elevated merger fractions (e.g., 2/11) for follow-up with high-resolution ALMA/JWST.

• Improvement: Implement a Bayesian multi-Gaussian fitting routine (e.g., using PyMC or a neural network) to automatically extract line fluxes, FWHMs, and redshifts from ALMA data cubes, including for faint detections (S/N > 2.5 per channel).

• What the improved AI can do:

• Fit all 18 CO(3–2) lines simultaneously, handling blended or asymmetric profiles (e.g., ID8.2).

• Estimate redshift uncertainties (e.g., z = 2.6655 ± 0.0008) and velocity offsets without manual inspection.

• Automatically detect serendipitous lines (e.g., the 92.737 GHz detection) and flag them as potential interlopers or new sources.

• Improvement: Train a classifier on r′–z′ colors and spectroscopic redshifts to identify cluster members (z = 0.75–0.772) and distinguish them from foreground/background galaxies.

• What the improved AI can do:

• Automatically identify the 9 confirmed members and predict additional members from the red sequence, even with incomplete spectroscopic coverage.

• Detect bimodal velocity distributions (e.g., possible merger in J0846.FG) and flag them for dynamical analysis.

• Estimate cluster velocity dispersion (σ v = 1446 ± 471 km/s) and dynamical mass (5.8 × 10 14 M⊙) with quantified uncertainties.

• Improvement: Train a CNN on VLA 6 GHz images to classify radio galaxies into wide-angle tails (WATs), narrow-angle tails, or compact sources.

• What the improved AI can do:

• Automatically detect the “C”-shaped WAT in J0846.FG and flag it as evidence of a merging cluster.

• Identify other WATs or radio relics in the field, aiding in cluster dynamical state assessment.

• Improvement: Use a neural network to fit modified blackbody models to unresolved dust SEDs, incorporating priors on dust temperature, emissivity index, and area.

• What the improved AI can do:

• Automatically estimate global properties (T d = 47.1 K, M d = 2.21 × 10 10 M⊙, β = 1.63) and their uncertainties.

• Compute µL IR and µSFR (39900 M⊙/yr) with proper error propagation.

• Flag when a single SED template is insufficient (e.g., for multi-source systems like J0846.BG), prompting spatially resolved follow-up.

Summary: These improvements enable AI systems to autonomously discover, characterize, and model strongly lensed protocluster cores, reducing human bias, accelerating analysis, and providing robust physical parameters for extreme starburst environments at Cosmic Noon.

Investigating the processes by which galaxies rapidly build up their stellar mass during the peak of their star formation (z=2-3) is crucial to advancing our understanding of the assembly of large-scale structures. We report the discovery of one of the most gas- and dust-rich protocluster core candidates, PJ0846+15 (J0846), from the Planck All-Sky Survey to Analyze Gravitationally lensed Extreme Starbursts (PASSAGES) sample. The exceedingly high total star formation rate (SFR) uncorrected for lensing magnification (mu) of mu SFR = 39900+23000-12900 M yr-1 is the result of a foreground cluster lensing at least 11 dusty star-forming galaxies between z=2.660-2.669, where the intrinsic value is estimated to be SFR = 5200+3200-2000 M yr-1. Atacama Large Millimeter Array (ALMA) observations uncovered 18 CO(3--2) emission-line detections, some of which are multiply-imaged systems, lensed by a foreground cluster at z=0.77. We present the first multi-wavelength characterization of this field, constructing a lens model that predicts that these 11 galaxies (mu 1.5-25) are contained within a projected physical extent of 280 times 150 kpc, with a velocity dispersion of sigma v = 246 plus or minus 72 km s-1. J0846 exhibits the rare case of a protocluster candidate whose core is strongly-lensed, offering a magnified view of the rapid stellar buildup within an overdense environment at Cosmic Noon.